Every feature

The left dock (system stats and settings), References, Assets, the centre dockers (Preview, Timeline, StoryBoard, Music, Stems, SFX, Image editor, ComfyUI, Prompt, Prompt chat, Console), the Queue and the Gallery.

New project, open, save, screenshot, record, recycle bin, reset project, settings and eject every loaded model, in one row that stays put whichever docks are shown.

Every list with 8 or more items opens with a search box: type to narrow it (word starts, then the letters in order), arrows and Enter to pick, a letter opens it searching.

Every docker has a button that is lit while it is on screen (open or collapsed); a click hides it, another shows it again.

Live readings of the computer, with a frame and graphs (temperature, VRAM) for each graphics card.

Each graphics card's watts (to its limit) and clocks, the CPU's clock (and power where the computer reports it), the energy since start.

Wh / kWh (graphics cards + CPU where it can be read), average and peak watts, on the job's log, its card and the project's job list.

Unloads every loaded model to free the graphics card.

The studio opens as its own window with a custom title bar instead of a browser tab; the browser still works.

The live log of ComfyUI, the workers and every job.

ComfyUI's own interface framed inside the studio, with reload and open-in-a-browser buttons.

Every button has a tooltip; Expert mode hides the help paragraphs once you know your way around.

Space pauses the canvas, the live preview or a video in the Preview docker, unless you are typing.

Click any thumbnail to open it full-window: slide through the list, zoom at the pointer, pan, play videos and sounds, open the original.

The preview dialog shows a picture's or clip's seed, CFG, steps, sampler, model, LoRAs, size, render time and prompt.

The Sketcher, Timeline and Image editor: docked, ⛶ fills the column's height and pushes the dockers below down; floating, ⛶ takes the whole window and closes the other dockers (back on restore). The Sketcher has its own resizable height in the column.

Every docker collapses to its header or is maximized; the others stay open at their own height, Esc restores.

A pin keeps a dock's height (or a side dock's width) as it is now; it can be made smaller but not taller.

Every dock's state, height, width, pin and maximize are kept in docker_status.json and loaded before anything paints.

Save a layout for each set of dockers you open together (sizes, which are on screen, the side docks and the floating panels), up to six screen sizes per set, loaded automatically.

The prompt docker floats as its own window; drag, resize, collapse, or dock it above, below or beside another docker.

Both are dockers of their own: float them, dock them as a tab in another docker, beside or inside a column, or in the left docker's list.

Make an empty floating docker with your own name and dock floating panels into it as tabs; it can float over the whole app or stay confined to one docker.

Show it with the statusbar's ▣ Docker or Ctrl+Alt+D, float main dockers with ⇱ and drag them in; an orange line shows where each lands, side by side or stacked, with splitters to size them.

Drag a main docker in by its own title bar; 📌 makes the ▣ docker a regular docker in the main column, 🗑 deletes it and its button; the statusbar button carries its name.

Right-click a docker you made (floating or permanent) for Rename, its layout, Make permanent / Float it again and Delete docker.

A named docker shows its dockers as tabs, side by side or stacked (⇆ / ⇅ in its title, or ⚙ > Layout); double-click the title to rename it in place.

Dialogs and message boxes open in the middle of the screen, or where you last moved them.

Where every floating panel is, its size, name and tabs are kept in panel_layout.cfg.

The StoryBoard docker can become a window of its own, dock as a tab in a named docker, or go back under the timeline.

Dark, Darker (OLED black), Slate, Light and High contrast change the whole app, image editor included.

Any colour for the buttons, lit tabs and sliders.

One slider scales every font and panel together, 60 to 175 %.

A colour for each part of the theme (background, panels, lines, text, OK, error, hover, canvas, selected).

Save a theme as a .json in the themes folder, load it from its card, export and import it; an import keeps the theme you had as <name>_old.json, saves the imported one at once, and ↶ Revert goes back (2.709.0).

Click any element on screen to pick it and change its font, size, colours, shadows, corner radius or icon, for this element, all like it, or its whole kind.

Dialogs and popup menus take clicks and keys as usual while the mode is on.

The image editor follows the app theme, or its own Pro, Studio or Classic look.

Everything lives in its own project folder: outputs, video, images, sound, characters and scenes.

The song and scenes, prompt, chat, characters, environments, references, assets, gallery, timeline and each docker's fields.

Copies the whole project under a new name and opens the copy; an existing name is refused.

The project you leave is saved first, then every docker starts empty.

Deleted files go to the project's recycle bin; restore them, delete them for good or empty the bin.

Empties the project and makes it again with the same name, stopping its renders and clearing its finished jobs.

The whole project as one XML file, self-contained with its media or as a small manifest, streamed so gigabytes are fine; opening one makes a new project.

New, Save and Load whole movies on the StoryBoard (cast, chapters, scenes and renders).

The project's images, videos and sounds; drop files on it to import them.

(2.635.0: a group is a FOLDER of the Assets folder) sort files into groups; Send moves the file into the folder, Add links it there too.

⛶ on the SolidCraft docker fills the whole window (title bar and status bar stay) and closes the other dockers; 🗗 / Esc brings the layout back as it was.

Out of the image editor's canvas-only Artist mode with Shift+right-click anywhere (no context menu); Ctrl+Esc / Ctrl+Shift+F still work.

Ctrl+Alt+S opens Settings from anywhere; Ctrl+Alt+Shift+T, ⟲ bitGen Default or ?theme=default puts the built-in default theme back (Dark, no element styles, UI size 100 %), ↶ The theme before undoes it; the built-in theme can't be overwritten or deleted.

📁 Project | 🌐 Global | All in the Assets dock; project assets in projects/<name>/Assets, global ones (every project) in projects/Assets; uploads go into Assets; right-click 🌐 Move to Global / 📁 Move to Project.

Tick, Ctrl-click or Shift-click tiles; delete moves them to the recycle bin, never a hard delete.

Dashed add tiles open the file dialog to add pictures, videos or sounds, several at once.

Pictures, videos and sounds that ride the next generation; drag them in from the Assets or the Gallery.

Every render, with tabs for images, videos and audio; drag from it onto references, the timeline or a scene.

One job at a time on the graphics card; pause, reorder, abort and retry, every job with its log.

Queued, running and interrupted jobs apart from finished ones, saved with the project; resume an interrupted job.

The running render shows as one card with its prompt, mode, size, progress bar and chunk / step / time.

Removes every done, failed and cancelled job from the list without deleting files.

Stop this render, retry it at the front of the queue, or stop everything.

Every finished job's details, with a button to put its seed back in Basic Settings.

Every video job shows which node is running and "step n of total".

Quick access, the app and project folders, every drive with its free space, an address bar, search with wildcards, details or large icons, new folder and rename; the same on Windows, macOS and Linux.

Every gallery tile has a save-as icon that opens the file dialog in Save As mode.

Save in the preview dialog copies the file to the last save-as folder without a dialog, never overwriting.

Saves the app window or the whole screen to the gallery.

Records the app or the screen as an H.264 mp4, with computer sound, microphone and a countdown.

A QR-code remote link: the phone takes photos and video, sends audio and talks to the model; everything lands in the project.

Any sound that plays fills the preview with a live visualisation (spectrum, wave, rings, particles, spectrogram).

A look-around view of any equirectangular video: drag to look, wheel to zoom.

One installer per system sets up Node.js, Python, ComfyUI, the node packs and the models, tuned for your graphics card.

The source with a README, the changelog of every release, contributing and security guides, issue templates and automatic tests on every push (scripts/ci-units.sh); docs/BETA.md says what is experimental and how to report a problem.

Downloads the models each feature needs (video, image editing, upscaling, 3D models, Viggle and more), with a Hugging Face token and the Hugging Face CLI for big files.

Installs the ComfyUI custom node packs; a missing one is named with its pack when a job starts.

Pick a model, start and stop its server, and rescan the folder; the running model is picked on start.

A live estimate of the prompt's tokens and a reply-budget slider.

Avoids most out-of-memory failures when mixing images, music and video.

Sage attention, Spectrum and Comfy Kitchen speed-ups per video style, with one-click installs.

Load a workflow, convert it to an API prompt, add your own toggles, dropdowns, sliders and text boxes and wire them to its values.

Run a wired graph from Generate with the controls you made; its load nodes fill themselves from the reference dock.

A UI workflow is turned into an API prompt (subgraphs and Set / Get nodes unfolded), now with ComfyUI's own frontend.

Model files a workflow names are matched to the ones your ComfyUI has, and the job log says so.

The engine for parametric 3D bodies and the Unreal Engine bridge settings.

The dropdown under it says what it makes: video, image, music, speech, a wired graph or a chat reply.

In Image mode a dropdown on the menu bar says what Generate does to the picture in the image editor, like the Krita AI plugin; Refine gets a strength slider and the button's words follow the choice.

With the image editor closed, Edit, Refine, Replace background and Compose use the References docker's pictures (Fill and Remove leave the list); with it open, they use the selected layer, or several picked layers; a chip says which.

One picture from up to three pictures: the layers picked in the image editor, or the References docker's pictures.

Ref (reference pictures), F/L (first / last frame), Text, and Swap.

/br/ on its own line splits the prompt into jobs that run one after another.

Enter sends the prompt to the local LLM, which rewrites it.

The image editor's AI, the timeline's Fill, path layers, AI scenes, the 3D look, characters, environments and VR worlds all take their words from the prompt docker.

What the picture must not have, sent with every generation; its status says whether the chosen model uses it, and the LLM's "Negative:" lines land there.

Save a prompt by its first line and pick it again from a list.

The running job's step count above the button and a progress bar under it.

Empties the prompt box only.

Speak records from the microphone and adds the words to the prompt.

Talk through an idea with the local LLM; prompts in its replies become blocks you can render straight away.

A chat is saved as "subject - date" and listed newest first; clear wipes the window and the model's memory.

Pictures in tray order with a one-line description from the local vision model, the videos and the sounds, so it writes prompts with the right tags.

With an LTX or Wan style picked the LLM writes for that model.

Spoken lines are put into H3's dialogue syntax with the language and the voice timbre; stray instructions are tidied away.

Speaks each answer with the chosen voice.

Asked for something longer than one clip, the LLM writes a numbered shot list that chains.

The LLM is taught equirectangular and side-by-side stereo prompt rules when you ask for them.

Mode, resolution, aspect, seconds, steps, CFG, seed, format, LoRAs, ControlNet and face fix.

-1 for a random seed, a number for that seed, a dice for random again, a button for the last seed used.

Extra LoRAs added on top of the style's for the next renders.

The picked video style with its model per phase, steps, CFG, encoder, VAE, LoRAs and negative; change a model there and it is saved into the style.

Basic Settings decide steps, CFG, seed, model and extra LoRAs for every video model (MiniMax H3, Wan, LTX).

A saved model set and workflow per mode, with phases for models that sample in stages; duplicate one to keep variants.

Reference pictures (up to nine), first / last frame, or text alone, with native sound.

A clip longer than one run is split and joined.

Low-VRAM attention and chunked feed-forward let a MiniMax H3 clip of up to 30 seconds render in ONE generation instead of joined 15-second pieces.

Install or update KJNodes, Yusu H3 Unified and RH DLSS 5.

An optional MiniMax H3 accelerator in a video style (On / Quality): about 2-3x faster than stock attention at its trained sizes, with a one-click install and its predictor in the model catalog.

Veda skips shots whose prompt describes a fight, a transformation, an explosion or a chase (it saved little there and softened the picture; ☑ Veda on action shots too keeps it), and replaces Long video's low-VRAM attention instead of being skipped; Spectrum keeps the first 2 and last 3 steps real on runs of 12+ steps so faces stay sharp.

Pages that fit the docker, folder tabs (SDXL, Pony …), a 📌 This project tab, resizable cards, an info page per model with its example pictures and prompts, and Civitai details + thumbnails + examples for every model kept in one shared model_info folder.

⤓ on any Civitai card downloads into the folder you picked once for that type (SDXL LoRAs, Pony models, Wan LoRAs …); Shift+click changes it; 📁 lists them (model_folders.json); 🔄 Sync fetches Civitai's details for every model on the computer.

A short description under every reference picture, video and sound, kept with the project and groups, used in the mapping lines, the prompt chat and ✨ Enhance; wall / floor / furniture / light picture types.

A dropped reference becomes a layer; ✨ Edit can take a reference as Picture 2.

Each Blender / Unreal object is matched to a reference picture, by its name or by hand. The first frame is made real with those pictures by an edit model, then LTX 2.3 3DREAL animates it along the render.

Settings ▸ Add-ons lists the optional engines (YuE2 music, Viggle native): what each is for, needs and size, and whether it is installed; ⤓ Install / ⟳ Reinstall / 🗑 Remove with the log in the tab.

Anything already installed says ⟳ Reinstall (a downloaded model ⟳ Download again) and fetches everything again; while an install or download runs its button is disabled and a status beside it shows the step, the percentage and the seconds, then ✓ or ✗ with the reason.

Settings ▸ General ▸ ⟳ Updates: check GitHub for a newer version (the new commits listed), ⤓ Update now (a git clone moves to GitHub's branch with your changed files backed up; a zip install gets the changed files), npm / pip when their lists changed, ⟳ Restart now; a quiet check after start.

The app folder holds start.bat (+ package.json, README, LICENSE); the server, the workers, the settings defaults, the setup scripts and the notes are in misc\; an install updated from a zip tidies its old copies away at the first start (quick prompts kept).

⛶ fills the room between the open Gallery / Queue / References / Assets / left docker.

Mark a reference picture as person, face, clothing, first frame or last frame; a first frame switches the render to first / last frame.

CFG above 1 adds a negative pass that strengthens only what your prompt adds.

Choose how sizes between H3's 32-pixel steps are handled: round down, or render larger and crop back exactly.

Each line carries its emotion and is spoken in the character's cloned voice; long voice clips are trimmed to an 8-second sample.

A language tag in the quotes speaks the line in that language.

A motion video drives the movement and a picture says who performs it.

Text, image, first / last frame, first / middle / last, lip sync, first / last lip sync, audio to video and video to video (style transfer); Auto picks by what the reference dock holds.

[0-2s] … lines in an LTX prompt give each stretch of the clip its own words in one render, untimed lines cover the whole clip; ⏱ adds the next line.

Switched on automatically for the modes that travel between pictures.

Text, image and first / last frame run on ComfyUI's own templates with only built-in nodes.

Each scene's card decides the mode, with or without the board's song.

Text, image, first / last frame and 3D render to real.

A render from Blender, a game or any CG clip becomes photoreal with the same layout and camera move; strong v2, strong or light variants.

Text, image to video, first and last frame, and a character reference (VACE).

BitGen builds the Wan graph from ComfyUI's own nodes and picks the right model files by name.

Wan's 16 fps is motion-interpolated to 24 fps to join your other clips; the reference dock's first sound is laid under it.

Up to three of each ride a reference render; missing slots are added to the workflow when needed.

A looping preview, a zoomable filmstrip and a selection box; Apply cuts exactly those frames.

Render a new part that starts on the video's last frame (or ends on its first) and joins it, always as first / last frame.

The playhead's frame from any layer becomes a first frame, last frame or reference.

Pictures are sent in order and named in the prompt from their types.

A name, details, a character sheet (front / side / back) and a voice clip; its @tag brings the sheet and voice into prompts.

Upload a photo or render and a reference sheet is generated.

Choose who is in the current scene.

Clothing entries follow the character; each keeps a dataset folder for identity models.

Locations as panoramas with a camera and a scene range, projected to a flat view for continuity.

Describe a story and the LLM writes the scenes, with the characters speaking their lines.

Type how many scenes, or let the project, your request or the LLM decide.

Ref (a new shot), First (exactly on a frame of the take before), Cont (that frame with the cast sheets), Trans (from that frame to the new location).

Rendering continues in the background while you edit.

Click a frame of a take to make it the next scene's first frame.

Watch, keep, redo, swap the character, or upscale a take.

Qwen Image 2.1 makes a scene's opening picture from the characters, a location and props.

Joins the rendered takes into one film in scene order.

The board in a dropdown, with a scene picker showing each scene's state.

"rendering step n of total" with a bar and time on each card.

Size, steps, LoRAs, accelerators, the upscaler and the tail frames.

Random for each scene or one seed for all; each card has its own seed box.

The StoryBoard setting, else Basic Settings, else the app default; the job notes say which.

A picture per scene from its prompt with any image model, using character sheets, locations and references; use it as the first or last frame.

Renders every scene picture that is not there yet.

The StoryBoard's latest message, in full on hover.

With H3_Viggle, each scene takes a motion video.

Its waveform opens auto-segmented; one scene per segment.

The writer uses the song's segments as scenes and makes the characters sing.

Every scene lip-syncs its own window of the song.

A timed transcript of the song says which scenes are sung and puts the words in the prompt.

Each take lands on the timeline at the time its part of the song starts.

Short transitions between song segments, rendered by the model or drawn as standard transitions.

A new person in existing footage; the performance, camera, background and sound stay.

Swap the character of a rendered scene into its next take.

For whole movies: load a video of any length, auto-segment it, swap each piece, keep and compile.

Full, Best or Fast.

The vision model and LLM write a swap prompt from the picture and the video.

A first frame with the new character carried through the clip.

Viggle-Animate through ComfyUI, or bitGen's own native engine that keeps the model loaded between segments.

Make up to four first frames per segment and pick one.

A black take is rendered again with everything that can blacken it switched off.

Swaps a person in 360 or 180-degree footage through a virtual camera; stereo eyes rebuilt from depth.

A performer inside a 360 or 180-degree picture: aim the camera, add the performer's pictures, sound and prompt, render with VR metadata.

Outside the performer both eyes show the picture itself.

Upscale a video to 720p-4K or by a factor, with colour match, temporal split and denoise.

NVIDIA DLSS 5 neural rendering on a video (default / natural / cinematic), upscaling to 1x DLAA, 1.5x-3x or 1K-8K, and optional frame generation up to 6x the frame rate; RTX only.

NVIDIA RTX Video Super Resolution on a video, with an optional Deblur first pass and a High Bitrate pass (preset A: VSR; preset B: Deblur, then High Bitrate; or pick the passes), scaled 1x-4x, a batch of frames at a time; RTX cards on Windows, through the DENO ComfyUI nodes.

DENO's R2V Turbo 8+2 recipe on your H3 renders: the text encoder unloaded before sampling, pass 1 in 8 steps at your size, a latent upscale (×1.0-2.0) and 2 refine steps with sparse attention, the sound kept; Auto turns it on for a card with 16 GB or less, and anything ComfyUI lacks is left out with a note.

A small chat panel like Facebook's: join with a user ID, ask in the shared Help room, add friends (they accept first), private chats, report / block, admins delete and ban; the server is PHP + MariaDB (community/ in the repo). a floating docker of its own (float / dock / tab into a docker, remembered - 2.711.0).

Joining asks for a first and last name and an email address (only the admins see them); a 6-digit code emailed to the address confirms it before the Help room opens (Send a new code, change the address); a ban also bans the email address, so a banned member cannot come back under a new user ID; older members are asked on their next sign-in.

A forum on tigerart.com.au with the same account as the 💬 Community: Help / Show your work / Ideas / General; members post at once, guests read and post (an admin approves a guest's post first); admins approve, refuse, remove and ban, and see real names.

Opens the upscaler for the last video you clicked.

A take you like sent through the StoryBoard's upscaler, or SeedVR2 at the upscale resolution; optionally every render.

Audio files and the sound of video files, kept as 48 kHz WAV.

Select on the waveform, set a fixed duration box, keep or delete the selection with click-free fades, undo.

The position marker at the start, Shift+click on the layer at the end: the range on that layer; ✂ Cut removes all of it from that layer only, ↶ / Ctrl+Z undoes.

Every edit to the timeline is a step: ↶ / ↷ in its menu bar, Ctrl+Z / Ctrl+Shift+Z over the layers.

Every cut is copied to the gallery; send it to the Sound assets.

Text typed, pasted or written by the LLM read aloud by OmniVoice or Ming-3.0 TTS, one voice or a cast (a voice and an emotion per speaker, from "Name (emotion): line"), sound tags, pauses; saved as WAV or MP3 into the Audio assets.

Neutral, happy, sad, angry, fearful, disgusted, surprised: Ming-3.0 TTS's emotion task (with cloned voices too); with OmniVoice the pace, a voice recorded in that mood ("Name.happy") and the sound tags.

The title, the date and time, the description, the text and the voices written into the WAV (INFO) or MP3 (ID3).

FLUX.2 Klein, SDXL, Pony, Illustrious, Z-Image, Qwen Image and more, each a saved workflow, model set and LoRA stack.

Generate Image launches ComfyUI when it is not running.

Model, VAE, encoder and LoRA fields list what ComfyUI can actually load.

Up to eight pictures per press.

One button sets the prompt docker to 🖼 Image + ✏ Edit and puts a restore instruction in the prompt (scratches, dust, stains, fading, noise and blur fixed; the people, faces and framing kept); a note says which picture Edit takes (the References docker's, or the image editor's layer); Ctrl+click starts the restore by itself, Shift+click starts a restore that also colourises a black-and-white photo (2.655.0).

Edit a picture from up to three references (FLUX.2 Klein edit) or more with Qwen Image 2.1 edit.

Black Forest Labs' 5.6B open model as three styles: distilled (fast), base (cfg 3.5) and edit (up to 4 reference pictures); the text styles switch to the edit graph when the reference dock has pictures.

Krea's 12B open image model as four styles: Turbo (8 steps) and Raw (52 steps, a negative prompt), each as safetensors or GGUF.

CircleStone Labs' 2B anime / illustration model as two styles (Aesthetic v1.1, Base v1.0; 30 steps, cfg 4).

Normal / Quiet (overnight) / Eco / Cool-down / Max performance / Custom: a GPU power limit, a rest between jobs, waits while a card is hot, throttling warnings.

Your models + LoRAs as cards (thumbnail, rating, release date, type, category, keywords; the description card on hover), select a model + LoRAs, override the style with them, ✚ a new style from one, ✎ edit a card, 🔍 look it up on Civitai.

Search civitai.red / civitai.com and download straight into the right ComfyUI folder with its card.

The price per kWh: every job's ⚡ card, its [power] line and the energy since start show the cost.

The tests a version needs, ticked as they are done (many by themselves), notes, 💾 a log to upload.

Graphics card (from a list), RAM and CPU when setup cannot find them; "Not right?" changes them.

Editor (no AI) / + small AI / + mid-size AI / Everything, picked for your card; what you'll be able to use (✓ ~ ✗, add later) and the download before you install.

The fixed daily charge on the bill: a job's cost adds its share for the time it ran.

Train a LoRA from your pictures with ai-toolkit (Ostris): nine base models (Z-Image Turbo, FLUX, Qwen-Image, SDXL, SD 1.5, Chroma, Wan 2.1…), trigger word, rank, steps, dataset with captions (or the assistant's), editable YAML, live steps / loss / samples; the LoRA lands in ComfyUI's loras/trained with its card; export the job as a zip.

A peak price and peak time windows (days + times, through midnight too); every other minute is off-peak; jobs split peak and off-peak energy.

Every job's energy, peak share, cost and remote GPU, day by day; ranges, totals, CSV.

Use another machine's ComfyUI, a RunPod pod (list / start / stop, its proxy) or an SSH tunnel; ✓ Test; Use it / Use this computer.

Your hardware, the package installed, what each package lets you use; ✎ Not right?; ⬆ Upgrade to a bigger package without setup.

A checkbox per model group, with its file count and size.

Every weight of a picture model (Lumina 2, Flex.1, FLUX.2 klein, SDXL) with ai-toolkit, parameter swapping for less VRAM; joined into one .safetensors in diffusion_models/trained.

The assistant on your own chats with Unsloth (Qwen3.5 0.8B-9B, Gemma 4, Qwen3.8 27B; LoRA / QLoRA / full); typed or imported (JSONL / JSON / CSV) examples; a .gguf in the LLM folder.

The vision model on pictures + answers (Qwen3.5 / Gemma 4 vision); the .gguf + mmproj in the vision folder.

A finished LoRA merged into its base model as a new checkpoint (ComfyUI), any card.

The FLUX.2 klein base 4B as a LoRA base model.

Ctrl+Alt+Shift+B: a screenshot of the window (✏ / ▭ marks), title, severity, where, what happened / expected, steps (⤵ the recorded clicks and keys - never what was typed), with the logs, system + settings (secrets hidden), page errors and failed requests; 💾 saves one .zip to send; earlier reports ⤓ / 🗑.

📤 Save and send puts the report on the website under Discussions ▸ Bug reports: as you when signed in to 💬 Community (shows at once), anonymously when not, or when ☑ Send anonymously (an admin approves it first); the public sees the title and your words, only the admins the logs, settings, system and screenshot; the admins set its status and download every bug as one list.

A style's .safetensors or .gguf model gets the matching ComfyUI loader (Load Diffusion Model / the GGUF loader) automatically - a GGUF style renders a .safetensors fine-tune and the other way round.

A style's diffusion model sitting in models/checkpoints (or another subfolder) is loaded through Load Checkpoint / the name ComfyUI lists, instead of failing "Value not in list"; a model in no folder gets a message saying where to put it.

A Workspace dropdown in the menu bar: Basic, Image editing, Video editing, Music video, Expert. Each shows the dockers that work needs, hides the buttons it doesn't (nothing disabled) and remembers your own changes; ↺ resets one.

One click makes a small test picture with every image style (Quick: 512 × 512, 4 steps; or Real) and shows a pass / fail table: the failing log line and what to do (a model ComfyUI does not list and which folder it belongs in, a missing node pack, ComfyUI not running, out of VRAM, a damaged file, a mismatched text encoder). ■ Stop, ↻ one again, ⤓ the report as .md.

A command palette over every docker, floating window, menu / statusbar button, Settings field, workspace, Generate mode, style, guide section and shortcut key; Enter shows / opens / runs it, and the guide opens in a 📖 Help window.

Text encoders, VAEs, LoRAs and ControlNets in another subfolder are used under ComfyUI's name; a file in no folder at all is named with the model catalog entry that downloads it.

Anything on the 🧪 test list the app sees working is ticked and kept (auto ✓), even after a restart.

Minimize a floating docker to its title bar, maximize it to the whole window and back, close it.

Drag it on the plan with the 3D view looking through it; arrows turn and tilt it, Shift / Alt raise and lower it.

Suggests the next test where you are (the image editor: blur, brush, eraser …), asks 'Did it work?' with a suggestions box when the app sees a test work, keeps your answers and suggestions in the log; off = nothing ticks or pops up.

A picture that finishes while the preview is open joins its strip; ✨ N new jumps to it.

With the image editor open, the image style dropdown lists only styles that can edit a picture, and puts your choice back when it closes.

Renders a first picture, turns it into a control image (lineart, depth, pose) and renders the final with a second model.

A two-stage style saved by name.

Re-draws each face of the final picture; an add-on since 2.593.0 (Settings ▸ General switches it on, its models download by themselves).

Face fix and ControlNet switched on or off in Settings ▸ General: off = hidden and never used; on = shown, and whatever they miss (node packs, models) downloads into the right ComfyUI folder by itself - also at every start; their Add-ons cards list every download link.

With the image editor open, Generate makes pictures and each lands as a new layer on the document in front.

Every picture in the Gallery and Assets can become a 3D model (Hunyuan3D 2.1 or TRELLIS.2).

Right-click a picture in the Gallery, Assets or References, or the image editor's canvas (the whole picture or the picked layer, also Layer ▸ Convert Layer to 3D), pick Hunyuan3D 2.1 or TRELLIS.2 and where the model goes: the image editor (a 3D layer you can orbit, render and generate from), the SolidCraft docker, or just the project's 3D models.

MiniMax Music3 or YuE2, with up to four songs per press.

A reusable caption (genre, BPM, key, mood, voice) with its workflow and settings.

Made in the Music docker.

The chat writes lyrics that fit the chosen music style and sends them to the Music docker.

Every song keeps its lyrics as a text file in the project.

Vocals, drums, bass and other.

Clone a voice from a short clip or a waveform region.

Plays the selection, or the whole clip when nothing is selected; ■ Stop, a playhead on the wave, a click clears the selection.

Speak a line in a cloned voice; OpenAudio or OmniVoice (600+ languages, voice design).

A second speech engine (inclusionAI, 0.5B, 44.1 kHz): cloned voices from a clip and its transcript, and designed voices. Installed from Settings ▸ Add-ons.

Generate with the speech mode.

Transcribe speech or extract timed lyrics from any sound or video, saved as .txt, .srt and .lrc.

Faster-whisper runs on this computer and each file is transcribed once.

Underdog's Bark 3.0 (163 MB, English only) for plain transcripts, voice clone transcripts, the phone remote and the 🎙 voice chat; lyrics, captions and other timed work stay on faster-whisper. Installed from Settings ▸ Add-ons.

Tencent's WeVisOne-VL (built on Qwen3-VL) as the prompt docker's vision model: it reads text in pictures, counts and finds things, and describes them for the LLM. Pick 4B or 2B in a dropdown; it downloads into your vision models folder.

See which llama.cpp build is installed and whether a newer one is out, and update it in place (the LLM and vision servers stop first; models and settings stay); the progress shows right there.

A file Facebook, Instagram and TikTok take as it is: H.264 High 8-bit, AAC 48 kHz stereo, a steady frame rate, faststart, 1080 on the short side at most, no timecode track; checked after the encode (each point ok or not, plus a note when a 9:16 is longer than 90 s). The social pack makes 9:16 Reels, the 4:5 feed, 1:1 and 16:9 at once.

Reels / Stories 9:16, Facebook feed 4:5, Square 1:1, Landscape 16:9 or the render's own shape: Aspect follows, and the finished MP4 is made ready for Facebook / Instagram / TikTok (the Export preset's settings, checked) before it lands in the gallery.

One toggle beside the seed: Random = a new seed every render, Fixed = the number beside it every time (typing a number or ♻ the last seed used makes it Fixed). In Basic Settings (video and image) and in every style (image, video, music) - a style can always render with its own seed.

While a project resets, a cover over the whole window says "Resetting Project..." with a spinning ring, running dots, what is happening now and the seconds; nothing behind it can be clicked, and it stays through the reload until the empty project is open ("✓ Project reset").

It asks the running ComfyUI where its model folders are and what its dropdowns hold, so models and LoRAs show even when they are not under the ComfyUI folder set in Settings (the Desktop app's own folder, a yaml on ComfyUI's command line).

Install, Update, Start (then Loading the model), Stop, Rescan, Scan folder and Save turn into a spinning "…ing" with the seconds, then ✓ or ✗ with the reason.

A transparent lyrics video with a bouncing ball and a line that changes colour as it is sung.

Pick a sound clip or asset, the whole song or the ruler selection.

Font, colours, outline, shadow, background band and the ball.

Correct the words speech-to-text heard, or paste the real lyrics.

Every word with its start and end, playable and correctable; corrections are kept for the song.

Move a word, or it and every word after it, earlier or later by a small or larger nudge.

The line being sung is lit and the word being sung brighter.

Docks beside the preview canvas so the song plays while you correct it; show the corrected words on the canvas.

The canvas with a toolbar on the left and a Layers panel on the right.

Drop videos, pictures or sounds on the canvas; each lands on a new layer at the position marker.

Select, move, scale by a corner and rotate by the top handle; Shift for 15-degree steps or one axis.

Click to add a text layer with font, size, colour, bold, italic, outline, shadow and alignment.

Each font list has the usual fonts under Common and every installed font under This computer (N); a font only in the app's fonts folder loads from its file.

Each font's name is written in that font (the search list too); the closed list shows the picked one in itself.

A text layer (image editor) or a timeline text becomes a picture; undo brings the text back.

The side handles slant from the opposite side (Alt: round the centre); text skews and stays text.

The visible layers kept as a temp copy, flattened (JPEG / BMP on white) and saved in the format picked beside 💾 into Images and the Gallery.

The frame round the text being typed hugs the letters and grows with each key; over a text the pointer is the text I-beam (elsewhere the new-text cursor); a click edits from between the letters where you clicked, also on wrapped paragraph text.

Pan, zoom and fit the picture.

Draw a box or paint a mask, then generate a picture or video that blends into it.

Every layer plays together with sound; loop the ruler selection.

Moved, scaled and rotated layers and text are compiled exactly where the canvas shows them.

With the Move tool the clip under the pointer is outlined in blue, the picked one in orange.

V1 as the base picture plus video, picture, sound, text, 3D and Karaoke layers.

Each layer has its own colour; the selected one is outlined.

Double-click a name, press F2, or use the Name box.

Move around at will; double-click a time to jump there and play.

Select a stretch, keep or cut it, trim the head or tail of a clip.

Arrow keys move a clip by frames, Shift by a second.

A project holds several timelines; switch, save and save as.

59 transitions, drawn as their shape on entry and exit.

Position, scale, rotation, opacity and text colour, tweened on the canvas and in the compile.

Linear, ease in, ease out, ease in and out, or hold.

Value and speed graphs with draggable keyframes and Bézier handles; Easy Ease, linear and hold in a click.

Blur a layer's movement (shutter angle, samples), hold it at a lower frame rate, or trail copies of it behind it.

Choose which source second shows at each moment of a video clip: forward, backwards, frozen, faster or slower, with keys on a curve, ◇ Key and ❄ Freeze at the playhead.

A click of the 🌫 tool makes animated fractal noise over the frame, a drag a linear or radial ramp; every setting keyable.

Push any layer's pixels with moving noise or with another layer's picture; the compile matches, video layers included.

A Shift+click of the 🌫 tool makes a particle emitter (point, line or box) with speed, direction, spread, gravity, wind and five particle shapes; every setting keyable, and the same picture every time it is drawn.

The bright parts of any layer streak away from a point, past its edges; video layers are baked whole by the page for the compile.

Pin a layer and drag the pins to bend it like a puppet; the other pins hold, and the pins can be keyed so the layer moves over time.

A grid of up to 8 × 8 points over a layer, dragged to bend the picture along smooth curves; keyable, and the compile matches, video layers included.

Key any property (opacity, scale, position, turn, generator numbers) to a sound's loudness, bass, mids or treble, from a quiet value to a loud one, thinned to the keys that matter.

Layers that draw the sound: its wave around the playhead (wave, line, bars) or its frequencies (bars, line, circle); every setting keyable; one clip or all sound.

The mix's loudness, bass, mids or treble (0..1) at any moment, for expressions on any property.

Press and move on the canvas while the timeline plays: the layer follows the pointer and the path becomes position keys, thinned to the ones that matter.

Random movement added to any property as keys (how often, how far, smooth or jagged, position x / y apart, a seed), riding on the move already there.

A property's keys averaged and thinned within a tolerance: fewer keys, the same shape.

Any layer becomes a card with depth and an x / y turn, seen in perspective by a keyable 3D camera (position, aim, zoom) with an orbit / dolly / track tool.

Cards blur by their distance from the focus, and a light throws the cards' shapes as shadows onto the cards behind them.

Each mask point can have its own feather (Shift+drag it with the Mask Pen); the edge glides between them, animated paths too.

Import CSV or JSON as data sets with a current row; {{set.column}} in texts, data() in expressions, and a column keyed onto any property a row at a time.

Label a layer's colour, opacity, position, scale or turn as a field; edit every field in one list, and templates offer them when dropped in.

The picture and text layers as a Lottie animation for the web and apps: moves, scales, turns and fades as real animated layers with their eases, effect layers as frames; checked against lottie-web.

Pick an Unreal project's folder and bitGen puts the bitGen Bridge plugin in (or updates it), switches on the plugins it needs and restarts Unreal with the project.

Width × height, size presets (4K to vertical and square) and a kept aspect ratio for the canvas and the compile; V1 is fitted inside and centred.

Click the subject (or box it) and it is followed through the clip (SAM 2, the AI matte or built in), refine edge, a see-through video layer.

Take a cut-out subject or masked area out of every frame, filled from the frames round it (+ a reference frame).

Solve the camera and the scene's points, pick points and place a layer flat, standing or facing the camera; the 3D camera follows the solve (with roll).

A ◇ beside every colour, effect, mask, text and time setting keys it at the playhead (then changing it keys it); the Graph Editor shows them and the compile matches.

Drag handles to resize an element, drag it to move it, hide it or change its text; every change is kept in themes/<name>.tmp, and leaving edit mode offers Save theme, Save theme as…, Reset theme, Not now or Cancel.

Double-click any words (buttons, docker titles, tabs, labels, headings, hints) to change them in place; rename a dropdown's options; change placeholders and tooltips; the floating tool panels are edited like the dockers; Shift+click (2.682.0) or Alt+click runs a button's action and opens sections / tabs / menus without leaving the mode.

A theme can hide an element only while Expert mode is on; Basic mode still shows it.

Ctrl+Shift+Alt+E styles an open dialog (Settings …) too; Ctrl+S saves the theme and leaves.

The side dockers fold to a thin title strip; the image editor's docker groups fold to their tabs and close with ✕; View ▸ Dockers shows or hides each one.

Follow a point, two points (scale and rotation), four corners or a flat surface (planar) through a clip to a fraction of a pixel, then apply it to a layer's position or pin a layer into it.

Draw a layer into four corners, tracked or dragged by hand; the compile matches the canvas.

Several masks per layer drawn with a pen (corners and curves), with add, subtract, intersect and difference, feather, expansion, opacity, invert and an animated path; pictures, text, shapes and video.

A small formula on a layer property (position, scale, rotation, opacity, anchor, shape and text animator numbers): wiggle, loops, delays, ranges, random, links to other layers; checked as you type; the compile follows it.

A position key's time follows the motion path for an even speed between its neighbours.

Ease a keyframe in and out, or one side only.

What position places and scale and rotation turn about; keyframable; the layer stays put when you move it.

The position path drawn on the canvas with Bézier handles to bend it; Smooth path in one click.

A picture, text, null or (since 2.495.0) video layer follows another layer's position, rotation and scale; it stays put when the parent is chosen or cleared, and the compile follows it.

Empty control layers to key and parent others to; never compiled.

Multiply, Screen, Overlay, Darken, Lighten, Color Dodge and Burn, Hard and Soft Light, Difference, Exclusion, Add and (since 2.495.0) Hue, Saturation, Color and Luminosity per layer, on the canvas and in the compile.

Show a layer only where the layer above it is (alpha or luma, or inverted); video inside text and the like.

Rectangles, ellipses, stars and polygons drawn with the Shape tool, with a colour or gradient fill and a stroke.

Trim Paths, Repeater, Offset Paths, Pucker & Bloat, Twist, Zig Zag, Wiggle Paths and Round Corners, every number keyable.

Draw a closed shape or an open line by hand, smoothed; Alt adds it to the picked shape layer as another path.

Several paths in one shape layer combined by add, subtract, intersect or exclude, holes included; the operators after it work on the merged outline.

Move, turn, scale, fade, space out, blur or recolour the letters, words or lines a range selects; Typewriter, Fade up, Pop in, Slide in, Wave and Blur in presets; every number keyable.

The letters run along an arc, circle, wave or a shape layer's outline, turned to it; the start is keyable.

Click a diamond on a clip for its opacity, volume, place, scale, rotation, colour and ease.

Drag along the clip; right-click, Alt-click or Delete removes one.

On sound clips too.

Blur, sharpen, glow, vignette, grain and crop move from key to key.

Trim and move everything after, move a cut, show other frames, or slide between neighbours.

Split at the position marker.

Split every targeted layer at once.

Play backwards or forwards at 2x, 4x, 8x; step a frame.

Previous or next edit point, the start or the end.

The edit tools and split are frame-exact on retimed clips.

Named, coloured flags with comments; next and previous marker.

Markers with a length, and markers that stay on a clip's frame.

HH:MM:SS:FF with a start and frame rate; drop-frame at 29.97 and 59.94.

10 to 1000 %, with the clip's length following.

Play backwards or ramp the speed along the clip.

Repeat, blend or optical flow.

A reversed or ramped clip's sound is rendered once and heard in step.

Teal and orange, warm film, cool night and more; exposure, contrast, saturation, temperature, tint and hue.

Import any 3D .cube LUT and blend it.

Lift, gamma and gain wheels; master and RGB curves.

Up to six hue ranges shifted, saturated or lightened.

Hue vs hue and hue vs saturation.

Waveform, RGB parade, vectorscope and histogram.

Set a clip's curves to match another clip.

The grade is baked into one table the canvas and compile share.

A phone's HDR video is tone-mapped into the movie.

Blur, sharpen, glow, vignette, grain, crop, green / blue screen key and stabilise.

Lens, mirror and warp (wave, twirl).

Smooth or locked, with a steady preview.

Colour and effects that work on everything under them.

An ellipse or box with feather and invert limits the grade and effects.

The mask follows the thing to the clip's end.

Voice clarity, denoise and gain, with a before / after preview.

On any clip's sound.

A clip's sound starts early or runs on late.

A video's sound as its own linked clip; unlink or bring it back.

Voice, music and sfx roles, ducking and loudness to a target.

A fader, pan, mute, solo and live level meter per track and for the master.

Every sound clip shows its waveform.

Everything that speaks is transcribed locally and placed where each clip sits.

Times, words, font, size, colour, outline or box, position.

Burned into the movie and / or an .srt beside it.

Choose words, cut them or keep only them; remove pauses and filler words.

This clip, its layer or the whole sequence.

Split a clip where its shots change, or drop markers there.

Sync angles by sound and cut live with keys 1 to 9; flatten into ordinary clips.

Another timeline compiled into one clip, re-rendered when it changes.

Small copies of heavy footage for playback; the movie uses the originals.

Shows which stretches play live, may drop frames, are too heavy, or are rendered.

Compile once and play the exact movie at full speed over rendered stretches.

Full, half or quarter size while playing.

Layers composited on the graphics card while playing.

Render a clip again in its own slot and switch between takes.

More seconds continuing from a clip's last frame.

Fill an empty stretch of V1, or a cutaway over a range.

Find a song's beats, then drop markers or cut a clip on them.

Draw a path and a subject (animal, person, vehicle) follows it, with size and pace.

A green-screen clip moved by keyframes, or Wan ATI making it travel itself.

An AI matte or a chroma key; every path layer is see-through.

Presets, pan, tilt, roll, height, dolly and lens keys, with camera, top and side views.

Drag on the canvas to pan, tilt, roll, change height and dolly.

Each video model is told the move it can follow.

A picture with depth flown along the path as a guide video.

Extruded 3D text or FBX, OBJ, GLB and glTF models, moved and keyframed like any layer.

Move, rotate and scale in 3D with G, R and S.

Material presets or your own, studio lights and your own point, spot, sun, ambient and sky lights.

Widget, snap, views, spin, reset, material, lights, key and bake.

A 3D layer is baked into a see-through video automatically.

A 3D layer rendered by Blender for the compile.

Rulers in picture pixels; drag guidelines out, move them or type their place.

Moved layers snap to guidelines, the picture's edges and centre, and the grid.

A grid over the preview canvas with its own settings.

Spacing, major lines, lines or dots, colours and snap distance.

Layered 3D text looks drawn live as you type; import your own from a PSD.

A text with a preset keeps it with ✨ Text Animators and 〰 On a path; 3D text takes a preset as its front face (sides in its colour).

The T tool's new text starts with the preset last picked; nothing selected = Properties ▸ New text ▸ Preset.

One character turns into another between two layers: write both characters, pick the two layers; the prompt is written, the first / last frames are taken, the video model renders it, it lands on a 🧬 Morph layer between them and layer 2 moves to start after it.

A timeline tool that selects single frames on one layer (click, drag for several, Shift+click, ← / →); Del / ⌫ cuts them out of that layer and closes the gap, ↶ puts them back.

✂ Cut / Del on frames through a clip joins the parts before and after into one continuous clip (a video / sound made into one "-join" file, frame-exact; a picture just shorter); one ↶. The preview loads the next clip a second early, so a cut never flashes black.

Tap Ctrl while the preview plays for a range's start and again for its end, as many as you like on the highlighted layer, each its own colour; drag a ▼ to move it (the preview shows that frame).

Click a ▼ to pick its range, Ctrl / Shift + click for more; ✂ Cut (and ⏩ / ✕) work on the picked ones, all when none is picked.

Every style keeps the Basic Settings you last used with it (misc/basic_settings.tmp): a style change saves the settings for the style you leave first, then puts back the new style's; the app opens on the last styles with their settings. A video style carries everything but the image set; an image style the image set.

Done green, installing a lighter green (with its time), warnings orange, errors red in the setup window's steps, log and downloads and in the install scripts; a quiet step says it is still working, why it can take long and that it has not stalled.

Install_all.bat / install_all.sh / install_all.command: Git and Node.js installed when missing (Windows: Python too), the repository cloned next to it, then the studio starts (the first start installs the rest).

A zip of PSD text effects (a pack) installs every effect in it as a preset in one go, with the fonts it carries; several PSDs can be picked at once too.

A Blender, 3D (Daz / Blender) or Unreal layer that is hidden or deleted closes its scene / live view in the image editor and the timeline (the Blender process ends, the Unreal View stops, Daz's waiting commands are dropped).

Cut every marked range at once (a clip cut several times stays one clip), or play each 2× - 10× faster to fast-forward a render.

The time under the pointer in a small box over the timeline; the preview plays at 0.125× - 8×.

A checkbox that puts the part after a cut or a Ctrl+K split on a new layer right above.

A text stretched to fill a shape layer's outline or flowed inside it (Fit), placed on it; follows an animated shape.

The shape filled with rows of text (font, size, colour, angle).

16 presets (Arc … Twist, Wedge) as an editable ⌗ mesh you drag with 📌.

Lower third, title card, end card and corner tag, or save your own.

The canvas tools and the selected layer's settings in a docker of their own, docked beside the canvas or floating.

A ribbon over the preview row with Create, 3D, Sketch, Solid and Unreal tabs (see CAD and 3D).

Nudge, song time, karaoke style and grids.

File, Edit, Image, Layer, Select, Filter, View, Window and Help, each item with its usual shortcut; the options bar shows the current tool's settings.

Tools with more in their group cycle with Shift and their key, or open with a right-click or long-press.

Type part of a command and press Enter.

Every key in one list.

The editor fills the space between bitGen's own bars; its status line moves into bitGen's status bar.

Tab hides every panel, Shift+Tab only the panel groups.

Every New or Open is its own document with its own layers, undo, selection, paths, channels and zoom.

The first document docks as a tab, the next ones float in their own windows; drag between the two.

Close and Close All ask about unsaved changes; tile, cascade, consolidate to tabs, next and previous document.

Color and Swatches; Properties, Adjustments, AI and Control; Layers, Paths, Channels, History, Actions and Comps; plus the Character, Paragraph, Sketch, 3D, Body and Unreal tabs.

Every tab group floats with its float button and docks back into the column.

Drop a floating group above or below another (an orange line shows where); the column's width drags from 200 to 640 px.

Dock floating groups as tabs in a docker you name.

X, Y, W and H of the picked layer with a proportion lock; rectangle corner radii; a vector shape's outline width and line style.

The last 80 steps; click one to go back or forward to it.

AI results, context, feather, megapixels, what Apply does, the look, rulers, panels and the options bar.

Solid colour or gradient fills, editable text and shapes.

16 kinds (Brightness/Contrast, Levels, Curves, Exposure, Vibrance, Hue/Saturation, Color Balance, Black & White, Photo Filter, Channel Mixer, Color Lookup, Gradient Map, Selective Color, Invert, Posterize, Threshold), live in Properties.

Ctrl+G groups, Shift+Ctrl+G ungroups; Pass Through lets a group's adjustments reach the layers under it.

Paint on the mask, Alt-click to view it, Shift-click to disable it, Ctrl-click to load it as a selection; density, feather, invert, apply.

A layer shows only where the layer below has pixels.

Fill opacity dims pixels but not styles; lock transparency, pixels, position or all; colour labels from the right-click menu.

Move layers up and down, Merge Down, Merge Visible, Flatten Image, Rasterize.

Keep the layers' visibility, position and appearance, switch between comps, update one, or go back to the last document state.

Del deletes the current layer (or clears a selection); Shift+Del always deletes the layer.

Esc puts the tools down (nothing picked); a click picks an object, Shift or Ctrl adds it.

With no tool, drag a dashed box: left to right picks the layers wholly inside it, right to left the ones it touches; Shift adds.

Shift-click picks every layer between two, Ctrl-click adds or takes one away; the Move tool moves them together and Del deletes them all.

With several layers picked, Ctrl+G puts them in one group (Ctrl+G on one group ungroups it), and the 🗑 button deletes them all.

A picture dropped on an open canvas asks: add it as a layer or open it as a new document.

In image-edit mode a generated picture always comes in as a new layer, with the style picked in the prompt docker (Krea 2 and every other style are listed for Generate).

A selection is filled with the prompt docker's style through its own graph (the picture encoded + a noise mask + differential diffusion), so it blends like the Krita AI plugin; Krea 2, Klein and the others can fill now.

Ctrl+T (or ⤡ Transform on the Move tool's bar) scales, rotates and skews (Ctrl + an edge); Paste drops the selection so the Move tool moves the pasted layer.

Free Transform's middle handles skew the layer (a skew cursor), Shift / Alt + a middle handle stretches.

More context around a selection, the selection pre-filled from its border, a soft mask, a start at 0.9 for models not trained to inpaint.

Generate at full strength, a small region rendered bigger, the mask sized by the selection, a blend past the edge with a colour match, Edit in a selection changes only the selection.

Generate / Remove / New background in a selection through the edit model (the selection painted green + Krita's instruction), so the new part belongs in the scene; ▾ Edit keeps to the selection, or to the layer's own pixels.

A style that cannot see around a selection makes its picture at the selection's own box, so what it draws is the selection's size.

SDXL / Pony with the Fooocus inpaint patch, Illustrious / SD 1.5 / FLUX.1 with inpainting ControlNets, Remove with MAT / LaMa first, Krita's context bounds and its Add content / Fill modes.

Every ✨ run queues without waiting; results wait in a docker inside the editor, ✔ Apply puts one above its layer in that layer's "✨ changes" group, 🗑 Delete drops it.

Drag a floating docker (a main docker floated with ⇱, a ▣ docker, Basic Settings, LLM tokens, the prompt docker, ⏳ Pending) by its title onto the image editor's panel column: docked there - fold, move, size or float it again from its bar; kept for the next start and shown / hidden with the editor.

"picture N" in the prompt brings the References docker's picture to Edit, Generate, Fill, Remove, Refine (through the edit model); "ref N" in Compose.

The node pack + the Fooocus / MAT / inpainting ControlNet files for the Krita-way fill, ticked and downloaded from Settings ▸ Add-ons.

In the floating band with the other floating dockers (not on top of all), outlined while a docker is dragged; Settings, References, Assets, Queue, Gallery and the StoryBoard drag in by their title bar.

Each folder filter dropdown keeps the folder you picked (folder_filters.cfg); Basic Settings are saved to basic.cfg when changed and read back at the start.

A floating ▣ docker dragged by its title bar docks above / below / beside a docker or in a side docker (the blue drop zones); Basic Settings, LLM tokens and the prompt docker drop into a ▣ docker and float out again.

Dropped on the seam between two columns (or at the window's edge) it becomes a full-height column with its own splitter; kept across restarts.

A StoryBoard scene interrupted by a restart takes its job back (running, its progress bar) when ▶ Resume is pressed in the Queue docker's Pending tab; videos and songs load paused until ▶ / Space.

The scene pickers' 🎬 Scenes tab (every scene's own pictures) and 🎞 Clips (any take's frames); ✎ First frame starts from a chosen Picture A (style, composition, poses kept) in a new or the global prompt's location, the instruction written by the LLM (🤖 Write).

Talk to the app hands-free, like a live chat: it hears you (faster-whisper, local, kept loaded), the local LLM answers out loud and does what you ask - the prompt, styles, Basic Settings, Generate, StoryBoard scenes, dockers.

Scenes as Model-manager-like cards in a grid with ➜ to the next, a click opens the full scene card; the LLM writes the scenes and every scene gets its storyboard picture as its starting (First) frame.

OpenMOSS's 5B image model as two image styles (SFT, Base); Settings ▸ Add-ons ▸ MOSS-Image downloads the model, text encoder and VAE you tick into ComfyUI and checks ComfyUI is new enough.

The Style panel lines text up (left / centre / right / justified, top / middle / bottom - buttons, flex rows / columns, grids and blocks of words), sets padding and margins for the element and for its words (each side, = for all four), and rounds each corner on its own.

🖼 sends a scene back to the LLM to write its opening frame's picture prompt, with the character sheets and the scene before's last frame (when it has a take) as references; 📁 changes the picture (another scene's picture, a take's frame, a project picture or a file); the picture fills the whole card in the grid.

Boxes on the canvas, each with what goes in it; filled one box at a time (✨ Generate per box), by the Edit model in one pass (the boxes outlined on a copy), or as a regional prompt in one pass (a new picture, each box's words only inside it).

A zone's 🎭 picks a character LoRA that works only inside that box (two different LoRAs at most), so two characters keep their own faces in one picture; the scene's words over the whole.

Pen pressure with a curve (presets or your own), lightest-touch / full-pressure window, the levels your pen sends, tilt, the eraser end and the side button; pressure → size / opacity for every brush; 7 pen brushes (G-pen, Mapping pen, Brush pen, Real pencil, Fineliner, Technical pen, Calligraphy).

Ctrl / Shift-click several clips in the Gallery, Assets or References and drag them onto the timeline together (oldest on the top layer, one after the other); ⛓ Sequential keeps the timeline's clips playing one after the other down the layers with no gap after deletes, new layers and trims.

A transition of N frames at every cut between video clips on different layers, no clip moved (the lip sync with the song stays exact - the clip before is held under the next one); one button mutes / unmutes every video layer so the full song plays alone.

The StoryBoard song strip auto-cuts on any grid (2-60 s; 29.9 = one Long video run per scene); locked segments stay; remembered.

ComfyUI updated in place (the portable build's own updater, or git + pip), Newest or Stable channel, a ComfyUI the app started restarted; a look at GitHub for bitGen Studio and ComfyUI after start, a ⬆ in the status bar when either is newer.

MiniMax H3's reference-to-video in 8 steps (FastVideo): a video style, a StoryBoard Workflow, an add-on that downloads it, and its recipe (shift 12/3, VSA, 8 steps) on any H3 graph that loads it.

A docker's title bar and any element with words beside buttons / boxes get new words (caret and buttons kept, a collapse keeps them).

Pictures (ComfyUI and sd-cli), music and video report their sampler steps live, and every docker, dialog and card that shows a running generation says step n of total.

Every pick starts on just that element; new words never spread to other dockers or elements.

Pictures / videos / sounds of a model with their prompt + seed / cfg / steps (read from the file or the render), a click opens the viewer, ✎→ to the prompt, 🖼 as the thumbnail.

A five-star editor in half steps, the model type (checkpoint, LoRA …) and an NSFW flag; the filter bar filters by both.

A file's prompt with its seed, cfg and steps to the prompt box + Basic Settings.

🔄 Civitai n/total while it syncs, ✓ synced with the counts when it finishes.

▣ docker membership, tab order, shown tab, hidden / collapsed, minimized floating panels, extension dockers - kept over a restart.

A model not found by its name is found by its file's content; the status bar shows reading n%.

The viewer sends a picture / video / song to a model's files with its metadata; Ctrl+↓ makes a picture the model's thumbnail.

The Character creator's 🧍 Pose sheet dialog shows thumbnails of the pictures in the pose_sheets folder; the picked one gives the character sheet its panels, poses and camera angles.

Open-source parametric 3D CAD (SolidCraft, MIT OR Apache-2.0) in its own docker, every SolidCraft feature; designs and parts kept in the project, SolidCraft's Save keeping them current; a picture of the model (or SolidCraft's render) into the image editor, the model as a timeline 3D layer, image editor sketch layers into SolidCraft, STEP / STL / OBJ / 3MF into the project.

The part to the image editor, the timeline, the 📏 Sketcher (2.630.0: a group - it was the House builder), Blender (opens with it, saved as a .blend - shown when Blender is found) and Unreal (a Static Mesh placed in the level - shown when an Unreal project is set; plugin 1.4.0); the docker shows only when SolidCraft is installed.

With no ComfyUI folder in Settings, the ComfyUI on this computer is found (environment, beside the app, home, Pinokio, Stability Matrix, drive tops, ComfyUI Desktop) and its checkpoints and diffusion models are listed in the Model manager and Basic Settings; Use this folder sets it.

The file, the files beside it and its model_info page, after a confirmation that lists them and the styles using it.

Every docker and floating window (extensions included) with where it is and ↺ Reset to its default place and size; ↺ Reset all.

Every group of ComfyUI's extra model paths: ➕ a new group (other_ui1, other_ui2 …, filled from the folders under its base), change / add / remove a path (📁 picks it), rename, is_default, 🗑 delete a group; the rest of the file kept, a .bak of the last; ⟳ Restart ComfyUI.

🔄 Sync / ⤓ Civitai info keep a model's video and GIF examples too; a video thumbnail plays on the card and its ⓘ page, the viewer plays it; a video can be a card's thumbnail.

✨ Enhance and the prompt chat write the selected LoRAs' keywords (trigger words) into the prompt; Enhance adds any the LLM left out.

A description or a picture -> CadQuery code (cadrille, or the local LLM with error repair) -> SolidCraft's own editable sketches, extrudes, cuts, holes, fillets and chamfers (checked by volume; else the STEP as one solid); the code editable and run again.

Bria's 8B model as two image styles: Edit Turbo (4-step instruction edits) and Dev (text to image + edits); up to 4 pictures <image_1>..; the w4a8 or GGUF Q4_0 files for about an 8 GB card; downloaded from Settings ▸ Add-ons ▸ Fibo 2.0 (CC BY-NC 4.0).

A team on a NAS folder: the leader approves members and their bitGen issues every member a signed key each day; everyone opens the same projects; a PC's owner lets members render on it (unrestricted or from - to) only while bitGen runs there and sharing is on; 🖥 Render on: This PC / a member's PC / ⚡ Spread, with the team's renders and progress. Same network or Tailscale. 2.589.0: an OPTIONAL add-on (its own zip - it may trigger virus checkers).

One prompt answered by each ticked LLM in turn, each reply in the chat under its name; ✓ Use this one makes it the answer, every round (request, every reply, the choice, the time) is recorded and 🏆 shows the best LLM for each job (screenplay, music video, song, image / video prompt, chat) with win %, average time and a CSV export.

🔁 Compare LLMs rounds keep the full prompt; 🏆 ▸ 🎓 sends the picked replies (examples) and picked-vs-not pairs into the trainer's LLM job; the trainer learns from Examples (SFT) or Preferences (DPO / ORPO, Beta) and writes a new .gguf.

The chat / writing LLMs in one dropdown (★ Qwen3.8 27B - the base for your own checkpoint, Gemma 4 31B / 12B, Qwen3.5 9B, Qwen3 4B / 8B / 14B), downloaded into the LLM folder; the trainer's Qwen3.8 27B base trains as a 4-bit QLoRA on a 32 GB card.

Any saved / API workflow: its picture / video / sound inputs counted, a preview made a save, missing nodes and model files reported, Prompt / Negative / Seed / Steps / CFG / size / length controls wired, the graph drawn; ▶ Test run with test inputs - saved into comfy_workflows/ + a graph when it works, ComfyUI's errors when it does not.

👁 turns a docker and its status-bar button off; 🗂 Layouts saves / applies / exports / imports the whole docker arrangement with the theme (layouts/<name>.json); ★ Save as the default = layouts/default.json + themes/default.json for a fresh install.

Krita's live mode: every change is rendered again a moment later (a whole-picture Refine at the live strength, its own style / steps / size / 🔒 seed), one render at a time with the newest change winning, the model kept loaded, front of the queue, never in the queue docker / gallery; ▣ on canvas with a mix; ✔ Apply as layer.

Only the canvas, full screen and frameless; Alt+P the colour palette at the pointer (square + hue strip, chips, recent), Alt+G the Layers panel + ⏳ Pending, Ctrl+P the prompt docker (generate from the layer you are on), Ctrl+B the brushes (presets, size, opacity, flow), Ctrl+Esc (or Ctrl+Shift+F) back.

MiniMax H3 reference mods for every project in refmod/<name>/: 7 types (character, location, vehicle, object, prop, flora, animal), media uploaded and sorted into images / videos / audio, ✨ autocaption with the vision model, 🧬 Extract (pictures + video frames) -> the .safetensors kept in the folder, @name in a prompt uses it.

A structure video (+ first frame, six-part prompt) -> the clip in 4 steps on MiniMax H3: the node pack + LoRAs as an add-on, its workflows imported through ⤒ Import & check.

An extension (ext/<id>/) can bring its own server code (ext.json "server"), mounted at start and shown only while it runs; Settings ▸ Add-ons installs one from its zip (📦) and removes it. The Team render farm is one - kept out of the install zips because it may trigger virus checkers.

Drag a tab out = it floats; onto another docker = a tab of it; onto the panel column's edge = a docker of its own there; ⠿ drags a whole docker; kept; Window ▸ Show All Dockers / Reset Docker Tabs.

The selected Unreal / MetaHuman layer's (image editor) or Unreal clip's (timeline, at the playhead) camera, live from Unreal; S / M / L, ⏸, ⚡ keep Unreal fast behind bitGen. Unreal must run, not be in front.

✨ Enhance and the prompt chat write only the keywords of Basic Settings' temporary LoRAs (Override style on) or of the selected style's LoRAs (off).

Shift adds, Alt subtracts, Shift+Alt intersects; feather softens the next selection.

All, Deselect, Reselect, Inverse, Color Range, Modify (Border, Smooth, Expand, Contract, Feather), load a layer's transparency or mask.

The selection shown in red; paint to add, erase to remove.

Click a thing to select it, Shift-click for more, or drag a box; runs on this computer after a one-time install.

The main subject selected in one step: built in (no download) or the AI matte (BiRefNet on this computer).

The subject put into the layer's mask, the background hidden.

Type what to select ("the person's head", "dog, car") and select it or make it a mask.

Split a layer into its parts, each on its own layer in a group.

A workspace with Quick Selection, Refine Edge and Brush, views, edge radius, smooth, feather, contrast, shift edge, decontaminate colours and output to a selection, mask or new layer.

The line snaps to an edge as you move along it.

Paint and the selection grows to the edges, with Auto-Enhance.

The sky reaching the top edge, soft along the skyline.

Save a selection into an alpha channel and load it again.

What a mask hides shows red while you paint it, or all the time.

20 brush presets (plus yours and imported .abr brushes), number keys for opacity, pen pressure for size and flow; a hard-edged pencil.

Tip angle, roundness and flips; shape dynamics, scattering, texture, dual brush, colour dynamics, transfer and smoothing, kept per preset.

Save the current brush as your own preset (brushes/<name>.json) and import the sampled brushes of .abr brush files.

Clone from a source; paint a layer back as it was opened.

The classic retouching brushes.

Content-aware or proximity match, filled from around the spot.

The copy takes the colour and light of where it lands.

Drag a selection onto clean texture; Normal mode with Transparent and Diffusion, or Content-Aware with Structure and Color.

Move or copy a thing; its old place is filled, with Structure and Color settings.

Click the red pupil.

Fill a selection from the rest of the picture, with no AI.

Choose the sampling area (auto, rectangular or painted), see a live preview, set colour adaptation and the output layer.

The paint picks up the canvas's colour as it goes.

Erase what is like the colour under the brush, or a whole area of one colour.

Linear, radial, angle, reflected or diamond.

Fill with the foreground, the background or a gradient, with tolerance, contiguous and all layers.

Drag a box, pick a ratio, Enter crops.

Move, scale, rotate, or type X, Y, W, H and the angle.

Corners along their sides, anywhere, in mirror, or a warp grid with presets and a bend.

3 x 3 to custom grids; split the warp crosswise, vertically or horizontally for more points.

Place pins and drag them to bend the layer naturally.

Which part goes on top where it folds; a pin's turn held fixed or left free.

Narrower or wider while faces, objects and edges keep their shape; protect the selection.

Forward warp, reconstruct, twirl, pucker, bloat, freeze and thaw masks.

Smooth, Push Left and face-aware sliders for the eyes, nose, mouth and face shape.

One click for scale handles, a double-click for rotate and skew handles; type stays editable and shapes stay vector.

Angle, rotate by 90, flips and W x H percentages round the centre.

A layer or group goes inside and stays editable; scaling, turning and warping lose nothing.

Open the inside, change anything, and every copy follows.

Replace the contents from a file, export them, turn them back into layers or plain pixels; New Smart Object via Copy makes an independent one.

Any filter or adjustment on a smart object is added as a re-editable filter: on / off, settings, opacity, order.

Paint where the smart filters show.

24 blend modes and an opacity per smart filter.

The contents live in a file in the project's linked folder; change the file and the layer follows.

Make an embedded smart object linked, embed a linked one, or relink it to another file; a missing file is flagged.

All 24 adjustments applied to the pixels, plus Auto Tone, Auto Contrast and Auto Color.

Local tone: lift the shadows and hold the highlights by amount, tone and radius; the HDR look with edge glow, gamma, exposure and detail.

Add or take away cyan, magenta, yellow and black in nine colour ranges, relative or absolute; also an adjustment layer.

Pick a colour on the picture and shift its hue, saturation and lightness; give a layer the colour of another layer, the merged picture or a file.

3D LUTs: the timeline's looks and .cube files from the shared LUT library, mixed by an amount; also an adjustment layer.

Blend another layer, the merged picture or one of their channels onto a layer with 11 modes, opacity and invert.

Image Size, Canvas Size with an anchor, rotation, crop to the selection and trim.

Blur, distort, noise, pixelate, render, sharpen and stylize filters, each previewed live; Ctrl+F repeats the last.

Click the picture to place a flare or swirl centre.

Surface, Smart and Lens Blur (depth, iris blades, bokeh highlights), Average; Field, Iris, Path and Spin Blur with click-to-place centres.

Reduce Noise, Dust & Scratches, Despeckle; Smart Sharpen that removes Gaussian, lens or motion blur and fades in the shadows and highlights.

High Pass, Minimum, Maximum, Offset, a Custom 5 x 5 kernel and HSB / HSL.

Displace (a map from a layer), Polar Coordinates, Pinch, ZigZag, Shear, Glass; Color Halftone, Crystallize, Pointillize, Facet, Fragment, Mezzotint.

Spot, point and infinite lights with a height texture; Fibers and Difference Clouds; Diffuse, Tiles and Trace Contour.

Cutout, Poster Edges, Watercolor, Dry Brush, Graphic Pen, Charcoal, Grain and Craquelure.

A photo developer: white balance, exposure, tone, texture, clarity, dehaze, vibrance, a tone curve, an eight-colour mixer, black & white, colour grading, sharpening and noise reduction, optics, geometry, grain and vignettes, with presets and your own.

Remove barrel or pincushion distortion, colour fringes and vignetting; fix perspective, angle and scale with a grid; fill the edges.

The view box over a small picture; channel histograms with statistics; RGB, HSB and CMYK under the pointer with colour samplers; every note listed.

Up to ten samplers, a ruler that straightens a layer or the picture, pinned notes and numbered count groups.

Four corners onto what should be square; Enter crops and squares the whole picture.

Grow or extend a selection by colour, select the sharp parts, move / scale / rotate the selection; defringe and remove black or white mattes.

Work on the active layer, only inside the selection when there is one.

Drop shadow, inner shadow, outer and inner glow, stroke, colour, gradient and pattern overlay, bevel and emboss, satin; copy and paste them.

One-click Gold, Chrome, Neon, Embossed, Glass, 3D Pop, Retro, Fire, Ice and Letterpress.

The letters follow a drawn line; move the start or flip the side.

Buttons in the Type options bar.

Eleven warps, text round a circle, along a path, or extruded 3D letters, all still editable.

Font, style, size, leading, kerning, tracking, scales, baseline shift, faux bold and italic, small caps, super- and subscript.

Alignment, justify, indents, space before and after, hyphenate.

Drag a box with the Type tool and the text wraps inside it; convert between point and paragraph text.

Make, apply, clear overrides and redefine styles.

Select letters while typing and change only their font, size, colour or any Character setting.

Ligatures, alternates, swash, ordinals, fractions, figure styles, small caps and stylistic sets from the font's own file.

Letters spaced from their shapes.

Convert text to a shape layer of its outlines, or create a work path from it.

Click inside a shape with the Type tool and the text flows inside its outline.

The same layered 3D text presets as the timeline, drawn live.

Warp, round a circle, along a path, vertical and 3D keep the text's preset (3D sides in its colour, darker).

Flow the text inside a shape (Fit sizes it), stretch it to fill the outline, or repeat it in rows as the shape's fill.

Shell Lower, Shell Upper, Fisheye, Twist and Wedge join Arc … Squeeze.

Drag the grid's points (presets, 3 x 3 … 5 x 5, split lines); the text stays typeable, the shape editable, the warp follows.

Corner and smooth anchors, close, cusp, 45-degree steps, auto add and delete.

Draw by hand and fit curves; add, delete and convert anchor points.

The curve bends smoothly through the points you click; preview the next segment.

Move, copy, combine and transform whole components; align and distribute them.

Anchors and handles, nudge, corner or smooth.

Draw a shape layer or the work path; combine, subtract, intersect or exclude.

Heart, star, burst, arrow, speech bubble, cloud and more, plus your own.

The work path, saved paths, shape paths and vector masks; fill, stroke, make a selection or a mask.

Reveal all, hide all, the current path; density, feather, invert, link and rasterize.

Crop, canvas size, image size and rotate move the paths too.

Round the picked corners of a path, a joined shape or a vector mask.

Cusp and smooth nodes, 15-degree steps, close, or carry an open line on.

The other curve tools.

Solid, dashed, dotted, dash-dot and custom lines, with start and end arrowheads (arrow, triangle, circle, square, diamond, bar and more).

Pick and drag nodes, handles and segments; add, delete, join, break, reverse, close and reduce nodes.

Colour, width, dashes, corners, caps, position, arrowheads, behind fill, scale with object.

A rectangle, ellipse, polygon or line becomes an editable vector line.

A line becomes a path that remembers its outline, and back again.

Rectangle with per-corner round, scallop or chamfer corners, 3-point rectangle, ellipse with pie and arc, 3-point ellipse, polygon, star, complex star, spiral, graph paper and common shapes, all live.

New, add, subtract, intersect or exclude a new shape into the picked one.

Preset, calligraphic, pressure and sprayer strokes that become filled shapes.

Straight, right-angle or rounded lines that follow the objects they join.

A 3-point callout with its text.

A closed curve or new shape is filled with the colour picked last.

With the Move tool the object under the pointer is outlined in blue and the picked one in orange; a click on empty canvas picks nothing.

After a shape is drawn the Move tool is picked with the new shape selected (can be switched off).

Position, rotate, scale and mirror, size and skew about nine anchor points, with copies.

Weld, trim, intersect, simplify, front minus back, back minus front, boundary.

To the active object, the selection, the page edge or centre.

Fill (uniform or gradient), outline and transparency of a vector shape.

Place layers inside a closed shape; they show only within it and move, turn and scale with it.

Work on the contents alone, finish editing, or take them out.

The Move tool takes the frame and its handles transform the contents with it.

Several pages of their own size in one picture, from presets or your own size.

Drag a new artboard, or turn a group into one.

Each artboard as its own PNG.

Parallel, horizontal / vertical and angular dimensions, with units at a dpi.

Dimensions snap to nodes, side middles, corners, outlines and guideline crossings, named as you hover.

Drag a dimension's end and its number follows.

Type a dimension's real size and unit and every dimension is measured with that scale.

Each dimension is a group of extension lines and a dimension line, redrawn from its measurements.

The picked object's overall width and height.

Every wall of a closed shape dimensioned outside, plus the overall size.

A dimension between every two points along a line, plus the total.

Pixels, millimetres, centimetres, inches or points.

Drag out of a ruler, move, lock, type a place, list and clear.

To the grid, guidelines, objects, nodes, midpoints, intersections and the page; the Move tool snaps the layer's box.

Spacing, major lines, lines or dots, colours and snap distance.

View RGB, Red, Green or Blue alone, or load a channel's lightness as a selection.

Saved selections you can view, load, rename and paint.

Painting, fills, filters and adjustments change only the picked channels.

An ink colour and solidity shown over the picture, merged into a layer when you want.

Default colours, swap (which repaints a plain background), large outlined swatches.

A strip of colours: click for the fill, right-click for the outline of a shape or type, or the foreground and background on pixels.

As the foreground a brush paints transparency.

Presets plus your own, stops with colour, opacity and location; used by the Gradient tool, the bucket, fill layers and shapes.

Define patterns from the picture, pattern fill layers, fill with a pattern, and your patterns in Pattern Overlay.

Record menu commands, dialogs, filters and Layers panel changes, then play them; sets of actions with a default set.

Step on / off, stop in a dialog, insert stops and menu items, function keys, button mode, load and save.

Play an action on many pictures or a folder, saved to the project, the gallery or downloads.

Played back at the same place, scaled to the picture.

If the document is landscape, portrait or square, has a selection, or a kind of layer, play one action, else another.

Save pictures as JPEG and / or PNG, resized, optionally after an action.

A tile you drop pictures on to run an action on each.

Load action sets from .atn files; matched steps play, the rest come in switched off.

A new picture in the background colour; save the visible picture into the project's Images.

PNG, JPEG, WebP, JPEG XL, AVIF, GIF, TIFF, BMP or TGA with quality, scale, a preview and the file size.

Title, author, description, keywords, copyright and URL, kept with the document and written into exported PNG and JPEG files.

Fit to the page, a size or a percentage, portrait or landscape, then the system's print dialog.

Pictures from files or the open documents in a grid on a new document, one layer each, with captions.

Each photo on a scan found, turned straight and opened as its own document.

Line the picked layers up on one of them: perspective, move / turn / scale, or move only.

Masks that blend layers into a panorama (soft seams, matched tones) or a focus stack (the sharpest layer at each spot).

Stitch overlapping pictures (files or the open documents) into a panorama document with one masked layer each.

Fuse bracketed exposures, lined up first, into one well-exposed picture.

Bitmap, Grayscale, Duotone (1-4 inks), Indexed Color (palettes, dither, Color Table with .act files), RGB and CMYK (a press model: paper, black generation, total ink).

See the picture as a press would print it and which colours it cannot print; on screen only.

Export the four printing inks for a print shop.

Open, place and drop .jxl pictures; export JPEG XL with a quality or lossless; .jxl pictures show everywhere in the app and are sent to ComfyUI and the compile as PNG.

Copy, cut, paste, paste in place, copy merged and system clipboard pictures.

Layers, groups, masks, opacity, blend modes, visibility, layer styles, vector masks and shapes come across both ways.

Embedded smart objects open as smart objects and are saved back as placed layers.

Export a path as SVG, place an SVG as a shape layer.

A page opens as its picture without its text plus one type layer per line of text; edit with every tool.

Add blank pages or another PDF's pages, move pages (buttons or drag) and take pages out; saved in the new order with untouched pages kept as they were.

Changed pages saved back with real, searchable text; unchanged pages stay vector.

Each line written in its own font from this computer or the fonts folder, so any alphabet stays real text.

Layers, masks, live text, shapes, fills, adjustments, styles, smart objects with their filters, paths, channels, the selection and layer comps kept fully editable, in a zip any unzip opens.

The model sees the selection plus a context margin and answers as a new layer inside the feathered selection.

Renders into the selection of the selected layer from the prompt docker; Apply blends it in.

Follow an instruction with Qwen Image 2.1 edit, typed in the prompt docker.

Rework at a strength, remove what is selected, or regenerate everything else.

Grow the canvas and continue the picture into it.

Make 1 to 4 versions, compare them, apply or discard.

Scribble, line art, soft edge, canny, depth, normal, pose or segmentation layers guide the AI through ControlNet.

Layers as references: put the thing from one layer into the scene of another.

The picked layers go to the reference dock for any render.

The reference docker is written to the project as it changes (closing the app keeps it), all its pictures are saved, and F/L, Swap or Text mode no longer drop pictures.

The StoryBoard's shared prompt is saved a moment after typing, when it is left and on close.

Generate inside a shape's outline or a layer's mask, clipped to it.

With a closed shape active the result goes inside it as a frame.

The Edit tab switches the prompt docker to image editing, the Generate tab to generating into the selection.

A new image made from the 3D view as you see it.

A CAD sketch of lines, rectangles, circles and arcs held in shape by constraints and dimensions, with a fixed origin and two construction axes.

Sketch Select, Line, 2-Point and Center Rectangle, Circle, 3-Point Arc, Point and Sketch Dimension, each with its own letter.

Click point after point; nearly level or upright lines snap and get Horizontal or Vertical; ends join where they meet.

Turn what is picked into dashed construction lines that are never drawn in the picture.

Pick and drag points, lines and circle edges while the constraints hold.

Blue is still free, black is fully constrained, orange dashed is construction, grey numbers are driven dimensions.

Every closed region becomes a shape layer above the sketch.

The closed regions filled, with the line colour and width you choose.

Back to the Move tool.

A .bgi keeps every constraint; moving or scaling the layer moves its sketch.

Coincident, Collinear, Concentric, Midpoint, Fix, Parallel, Perpendicular, Horizontal, Vertical, Tangent, Equal and Symmetric.

How many are left, with a tick when fully constrained.

A constraint already set by the others is refused, and a dimension becomes a driven one instead.

A line's length, two points' distance, two lines' angle, a circle's diameter or an arc's radius; type the value, sums work.

Type a length and angle, width and height, diameter or radius as you draw; the values are kept as dimensions.

A centre and a corner, the number of sides, kept on a construction circle.

Two half-circles joined by tangent lines.

A circle from a diameter or three points.

A smooth curve through every point you click.

Cut away the piece of a line, circle or arc between the nearest crossings.

Run a line on to the next line or circle.

A tangent arc or a straight cut at a corner, with its value as a dimension.

A copy at a distance that stays parallel or concentric.

Click a chain, then the pointer picks the side and distance (or type it); lines drawn any way round go out or in together and rounded corners stay joined.

Mirrored copies held symmetric about a line.

Copies in rows and columns or round a point, with the same constraints.

Lengths, angles, radii, areas and perimeters of one item, between two, or inside a closed region; copy the numbers.

Pixels, or millimetres and inches with a scale.

An R12 DXF that CAD, laser and CNC programs open, with construction lines and dimensions on their own layers.

The sketch at a real size with its lines and profiles.

Lines, circles, arcs, polylines, splines and ellipses into a sketch, fitted or at real size, ends joined into profiles.

Tabs over groups of big buttons; each group's name opens every item in it; fold it down to its tabs.

Create (new sketch, lines, rectangles, circles, arc, polygon, slot, spline, point), Modify (fillet, chamfer, trim, extend, offset, mirror, patterns, construction), the 12 constraints, Inspect (dimension, measure), Insert (DXF in and out, SVG out) and Finish Sketch.

Select (the Move tool, pick nothing), Draw (the shape tools) and the after-drawing switch.

Picking a sketch tool or sketch layer opens the Sketch tab; a 3D layer opens the 3D tab on the timeline.

Create (text, 3D text, 3D model, camera, path, the canvas tools), 3D (widget, views, key, bake) and Sketch.

The frame at the playhead, or a blank picture of the timeline's size, opens in the image editor with a new sketch layer.

The toolbar can be kept on or left to show only when needed.

The CAD toolbar shows by itself while a sketch, a sketch tool or a 3D body is picked; the ⌬ CAD button in the tool column keeps it on (the timeline has its own).

Select leads the Sketch tab, Create Sketch hides while a sketch is edited, Finish Sketch goes on to the Solid tab.

A ＋ beyond each end of the picked line starts the Line tool from that end.

A 3D body drawn from its features as a smart object, so moving, scaling, filters and masks all work.

A sketch's closed profiles by a distance, one side, symmetric or two sides, as a new body or joined, cut or intersected with one.

Extrude and Revolve let you click the closed regions to use, highlighted under the pointer.

Profiles spun round the sketch's X or Y axis or one of its lines, by an angle.

Round or bevel all, top, bottom, vertical or horizontal edges.

Hollow the body, open at the top or bottom.

Cut by a plane; one side stays, the other becomes a new body layer.

Named values and expressions with units (mm, cm, m, in, deg) used by every feature; change one and the body rebuilds.

Every feature in order: suppress, edit, delete or reorder, with errors in red.

A sketch edited later rebuilds the bodies made from it.

Home, front, top, right, left, back, or drag the body round.

Click a face or corner of the cube for that view, drag it to orbit; Shift + middle-drag orbits a body, middle-drag pans, the wheel zooms.

Colour, metal, roughness and edges, and the view's turn, tilt and roll.

The body as a file for printing, 3D apps or CAD.

Send a body to the timeline as a 3D layer; it updates by itself when the body changes.

Open the image editor with that body picked.

OpenCascade running in the page (exact fillets, booleans and shells, nothing to install) or Blender on this computer.

A Daz or Blender scene in a picture, drawn again by Blender whenever its view changes; masks, filters and blend modes work.

A .glb / .gltf / .obj / .fbx / .stl from the project's 3D models opens as a 3D layer: Blender puts it in a lit scene with a camera framed on it.

The scene camera, front, back, left, right or top, or turn, tilt, distance and lens; drag on the canvas to orbit, pan and zoom.

Pick a figure and bone and turn it on top of its pose.

Pick a frame of an animated scene; render it Quick, EEVEE or Cycles, with a transparent background.

Move, turn, scale or hide the scene's objects, by numbers or by dragging them on the canvas; the .blend is not changed.

With a 3D layer picked, the prompt docker's Edit and Refine render the scene with Blender and work from that picture; the results come into the AI panel, and ✔ keeps one as a layer.

The timeline's Blender scene tools over the image editor's canvas (open a .blend, scene camera or free view, move / rotate / scale objects, save); the view becomes a layer, and Edit in the prompt docker makes a new layer from the scene.

Name a References docker picture in the prompt ("the face of the woman in picture 1") and it goes to the edit model with the scene; the numbers and the word are changed to the model's own (image 2, <image2>, Picture 2).

↺ Built-in lists the built-in image or video styles that are missing and puts them back; the Krea 2 styles come back once on their own.

🔍＋ / 🔍－, Alt+drag (forward = in, back = out) and ⌖ Zoom to object (Z); in the image editor the player bar and Auto-key are left out.

A click on a skinned body picks its armature, so moving it moves the figure (not just the widget).

Hunyuan3D 2.1 (shape) or TRELLIS.2 (shape and texture) make a .glb from any picture.

The local LLM plans a 3D scene from 58 kinds of object, a camera path and 3D models for the rest; Blender builds it.

Quick 3D modelling in bitGen: draw shapes on the ground or on faces and push / pull them into solids; models kept in the project, saved as you work, undo / redo.

A face becomes a solid; the top of a box makes it taller; a shape drawn on a face goes in (a pocket) or out (a boss); pushed to the far side, a hole through; a line across a face splits it.

Type lengths (m, cm, mm, feet, inches), two sizes, sides or degrees at any moment; the pointer snaps to corners, midpoints, edges, faces, the origin and the red / green / blue axes (arrows or Shift keep one).

The drawing tools; Offset makes a loop inside a face (walls round a room, frames).

The editing tools; Ctrl+Move copies; groups are picked as one and edited inside with a double-click.

Iso / Top / Front / Right / Back / Left, perspective or parallel, everything in view, shaded / monochrome / X-ray / wireframe, shadows.

A picture of the view into the image editor, a 3D layer on the timeline, a scene in Blender, a mesh in Unreal; .glb / .obj / .stl export; project 3D models imported as groups.

A room is drawn with two clicks (walls, floor, ceiling); dragged near another room its wall snaps onto the other's and the two share one wall; each wall face takes the colour of the room it faces.

Free-standing walls (closed into a room when they meet their start), floors without walls, a room's ceiling on / off, straight stairs.

Single, double, sliding and glass doors or a plain opening; one-pane, two-pane and sliding windows with a sill; a click on any wall (through a shared wall too), dragged along it, sized in the panel.

Gable, hip, shed or flat over one room or several, with pitch, overhang and thickness; the gable ends walled up.

Downlights, ceiling-mounted lights, pendants, chandeliers, spots, wall lights, floor and table lamps, each a real light with its fixture, colour (warm to cool) and brightness; 🌙 Night shows what they light; exported as glTF lights.

Whatever is picked, or the next one the tool makes: sizes, heights, styles, colours, duplicate, turn, roof, delete.

Every room with its walls, floor, ceiling, skirting, cornice, doors and windows (Room 1 ▸ Wall 1, Ceiling 1 …), roofs, floors, stairs, lights, furniture, cameras; 👁 hide / show; room groups; skirting + cornice on new rooms; a wall dragged across its line (Alt: freely), Shift+click = the whole room.

A picked camera or piece shows yaw / pitch / roll rings to drag (5° steps, Shift free); Turn / Tilt (Pitch) / Roll in its panel; a camera's Lens presets (Fisheye 8 mm, Ultra wide 14, Wide angle 24, 35, 50, 85, 135, 200 mm telephoto), Focal length and Field of view linked.

View ▾ lists the cameras; one picked = the view from it (its lens), following its panel; an orbit leaves it.

Open (%) and which way: doors swing in / out, windows' sashes out / in, sliding ones slide a pane.

Each new room with its roof and a floor slab under its walls; per room on / off.

Standing / sitting / walking / waving mannequins placed on a floor, named, sized.

A house saved by the old House builder becomes a Sketcher model (rooms, openings, furniture, people, cameras, lights, stairs).

Materials (brick, stone, timber floors, tiles, marble, carpet …) with their textures, painted on faces, walls and floors.

A camera placed with two clicks; look through it; its shot + depth + normal maps into the Image assets; a 360° panorama from it as a Location.

The LLM (or a room's usual set) fills a Sketcher room with furniture.

The furniture library (39 pieces: living, dining, kitchen, bedroom, bathroom, lights, outside; its materials and textures) placed on a room's floor, turned with R, dragged, sized, coloured, duplicated; the project's 3D models brought in as groups.

⛶ closes every other docker (the title bar and status bar stay); ⇲ Dock opens them again.

(2.627.0: a clean render + its depth map, the prompt led by "Turn this 3D render into a photorealistic photograph of a building ...") the Sketcher with the prompt docker open: the prompt docker goes to Image with an edit model and "create a real photo of the model in view"; Generate turns the Sketcher's view into a photo, filed as an ordinary image in the Image assets; the mode, style and words come back when the Sketcher closes.

→ Image editor sends the view as a 📏 Sketcher layer (sent again, it is updated); → Timeline sends the model as a 📏 Sketcher layer (3D).

Click a face: Extrude (+ out, − a cut) and Twist in the panel; drag it to extrude with the pointer (Blender-style); Ctrl+drag turns it and twists the extrusion; type a distance to change the last one.

Drag an edge or a corner to move it (Shift: one axis); Shift+click more edges, Ctrl+click the loop round a face; an edge's Length, Extrude edge (a new face) and Fillet (radius, pieces); a corner's X / Y / Z.

Grips on a picked door / window's sides, top and bottom resize it; a window dragged up / down; Elevation (height from floor) in its panel.

The Room / Wall tools' Offset from floor edge (the outside face that far in from a floor's edge); drag one wall of a room (the room stretches) - the floors' snap regions show and it snaps onto an edge; rooms snap to floor edges too.

The ground round the building: raise / lower / smooth / flatten with a soft brush, draw contour lines at their heights and the ground is shaped smoothly between them, a heightmap picture (black = lowest, white = highest), level the ground under the rooms; its size, detail, material, contour interval, earth sides; contour lines on the ground and on the floorplan.

A level pad cut into / filled onto the ground: its corners clicked, its level (the ground's mean, or typed), batters at 1 : n back to the natural ground, cut and fill / cut only / fill only, the cut and fill volumes in m³; picked to re-level, re-slope or apply again; on the floorplan.

A rectangle (two corners or typed) or free-shape pool: shallow / deep end with a sloping floor, edge tiles (coping), decking or both, their materials, the lining, the water; the ground cut away under it and levelled round it; its volume of water; on the floorplan.

Click points along the way, the driveway curves through them; drag its points and Bézier handles to shape it (Alt: one side), Delete removes a point; width, surface material, kerbs, on the ground or at a level, smooth the curve; on the floorplan.

The Eraser / Delete / 🗑 Take this wall out removes just one wall (the floor and ceiling stay, a shared wall goes from both rooms so they open into each other; ↺ Put back); each wall Auto / Interior / Exterior, an interior wall's outside face in its inside colour, interior walls lighter on the floorplan.

A docker of wall types made of layers outside to inside (brick, air gap, wrap, timber / steel frame with batts, insulation board, plasterboard, render, cladding, block, tiles ...), each layer's thickness, the total wall thickness and its cut drawn to scale; used for a wall, a room, every exterior or interior wall - the walls are built that thick, coloured by their outside layer, the seams on their tops and the floorplan.

Each wall ends square at its corners (the corners and free ends are pieces of their own); drag out of a side of a corner or a wall's end piece with Select and a new wall comes out that way (from its top: any way), with the wall's height, thickness and type.

Double-click a wall to split it (3D or floorplan); pick a piece to delete it or drag it out, new walls joining it to the room; the Wall tool starts / ends anywhere on a room's wall; a corner's / wall end's width (a pier); the clicked face is the sketch plane the drawing tools draw in (2.687.0: nothing drawn over it), ⊥ Look at faces it; 2.688.0: a click on a wall's end piece picks the whole wall, a double-click pushes / pulls the wall from that end, Ctrl+drag holds a move horizontal / vertical, the bar stays on screen, ✨ Photo's prompt names only what the model has.

Beside the prompt docker's image / video style dropdowns: toggle between the current and the previous style; with Override style on, its Basic Settings values and temporary LoRAs come back too; kept in model_settings.tmp (2.683.0: works with the app's own style change; — none — / — template — count).

PrunaAI's image model on MiniMax H3: the P-H3 Image style (text to image, 4-8 MP) and P-H3 Image Edit (edit a picture with up to 7 references, <image1> is edited); its own prompt upsampler turns a short request into the full prompt. Settings ▸ Add-ons ▸ P-H3 Image downloads its LoRA, text encoder, VAE and upsampler (~17 GB) onto the MiniMax H3 FL2VA transformer the video already uses. Needs ComfyUI with the P-H3 nodes (master / nightly, 0.44); MiniMax H3 Community License.

The image preview's look dropdown + 🎨 Change style (2.681.0: the look stays picked) makes the picture shown again as 📷 Photoreal, 🧸 Pixar, 🧊 2.5D, 🌸 Anime or 🖍 Cartoon with the prompt's image style (an edit model follows an instruction, keeping the picture; an SDXL / FLUX style refines the whole picture); with Override style on, Basic Settings' steps, CFG, seed, checkpoint and LoRAs ride along; the result opens in the preview; a progress bar under the picture shows its steps (2.677.0); ⇆ Before / after shows the original and the result side by side (2.679.0). (2.685.0: an SDXL / Pony / Illustrious style remakes it with its own workflow, so Override reaches it; the bar says what Override applied) (2.686.0: a style that is no edit model draws the ✏ line art first, then renders the look from it through a line-art ControlNet of its model family).

Draw roads like SimCity: Dirt road, Street, Road, One-way, Avenue and Highway (lanes, footpaths, medians, barriers, markings), following the ground; steeper than the type takes, ⛏ cut the ground to an even grade or 🚇 tunnel through (a bridge over a deep valley); junctions found automatically with traffic lights, stop / give-way signs, stop lines, crossings, street-name and speed signs; drive on the left or right.

A roof's gutter is a profile: Quad, Half round, OG (ogee front) and Square line built in, or your own drawn in the profile editor like a skirting or cornice (lines and Bézier curves), swept along every eave with its downpipes.

Skirting, cornice, frame and sill profiles drawn with nodes and Bézier curves (the built-in arcs one curve each); double-click a segment to add a node, right-click to convert a segment to a curve / line or delete a node.

⤓ Update ComfyUI first runs misc\clear-comfy.bat (every ComfyUI process, port 8188, its caches); on exit the same, every bitGen Studio process ended and the terminal window closed.

Every roof's eave as Simple, Open (rafter tails), Boxed (a flat soffit) or Closed (a lining on the slope), fascia + barge boards, quad or half-round gutters with downpipes, their own colours.

The Sketcher's objects panel lists every shape (solids, shapes, lines, groups; click to pick, drag a face to push / pull - a group's too), a Construction heading with planes (kept from the sketch plane, or ground / front / side) and construction lines (Tape measure + Ctrl, snapped to), and renames any row by double-clicking it.

Gable, hip, skillion, Dutch gable, clipped gable (jerkinhead), gambrel, mansard, saltbox, pyramid, butterfly, flat and follow-the-walls, picked from pictures, each with its own pitch / proportion / flip.

Floor plans for every storey, four elevations and a roof / site plan drawn to scale on A4 - A1 (landscape or portrait), all on one sheet or one view per page, with dimensions in mm, levels, room names and areas, and a title block (north arrow, scale bar, sheet n of m).

The Sketcher's own image edit style for ✨ Photo, beside the button; remembered.

Pre-painted steel (corrugated / ribbed / standing seam), concrete or terracotta tiles, slate, shingles, polycarbonate: colour, pattern, a typical minimum pitch warning, their weight in the framing checks, named in ✨ Photo.

Rafters, ridge, hips / valleys, break purlins, ceiling joists, underpurlins + struts (rafters and ceiling joists checked, ✔ Size them), or a truss layout, on every pitched roof type.

Timber or light-gauge steel frames in every wall: plates, studs at 300 / 450 / 600, jamb + jack studs, sill trimmers, noggings, nominal strap braces, corner posts; totals per room. Stud sizes not checked (AS 1684's stud tables).

Slab on ground, waffle pod, stiffened raft, suspended slab on piers or a raised timber floor (stumps, bearers, joists) under a room; joists, bearers and door / window lintels checked by beam sums from the grade's properties (✔ Size them); 🪵 Show framing. A guide - confirm against AS 1684 / span tables, AS 2870 for slabs.

A new wall's end stops against the wall it meets (no cap of its own); end to end the old wall's end is shaped round the corner; on a wall's line it is cut to that wall's face. Wall thickness remembers the last one set (over 2 = mm).

The Wall tool pulls a point onto another wall's end or corner when it is close (⊕ joined; Shift = free); joined walls stay together - drag one and the other's end follows, so it grows or shrinks.

Built-in robes, walk-in robes (L / U), linen, pantry and broom cupboards, bookshelves, shelving: a dialog for the layout, sizes, bays (hanging, double hanging, shelves, drawers, shoes), doors, handles and colours with a live front view and plan; 💾 your own templates for every project; placed against a wall facing in, ↔ Fit the wall; drawn on the floorplan and plans.

A modelled female mannequin (by dimasalcalde and zahers, CGTrader) for People in the Sketcher, scaled to the person's height.

✨ Photo sends the render, its depth and a labelled board of the model's material and background pictures to the edit model, so the photo uses them.

Each wall's inside and outside, a window's frame and glass, floors, ceilings, trim, roofs, stairs: a colour, a built-in material or a picture (from the Refs docker, the project or a file) as a repeating texture (size, turn); the paint bucket paints one wall side (Shift: all on that side).

A reference picture behind the model (filling the view) or wrapped round the scene as a panorama sky.

The picked thing is tinted in the theme's accent, the one under the pointer blue (not when it is picked) - in 3D and on the floorplan.

Forward (up) zooms in, back (down) out, round the point pressed; a plain right-click still opens menus.

⊡ fills the view with what is picked: the Sketcher's faces or building parts, the timeline's picked layer, the image editor's selection (or the picked layer).

The Sketcher's Select and the timeline's and image editor's Move are drawn as a pointer arrow.

A click picks one thing, Ctrl+click adds more (rooms, walls, doors, furniture, lights, cameras, faces); drag one to move them all, Delete erases them all. Esc in Orbit = Select; Esc never docks the maximized Sketcher.

The Sketcher seen straight down in the app's theme colours, one storey at a time: walls cut solid, doors with their swing, windows with glass and sill, stairs with treads and UP, furniture and lights as plan symbols, each room's name, Width × Length and area; every tool draws on it and a click picks a symbol; Orbit turns it flat.

One Esc drops the tool and the pick and goes to 🔄 Orbit in every Sketcher view (3D, floorplan, a camera's view).

Every room's panel has Width and Length (a rectangle keeps its corner, another shape is stretched); the Room tool's Width × Length places a room that size in one click.

In 🎨 Artist mode Alt+G shows the Layers panel and ⏳ Pending over the canvas and Ctrl+P the prompt docker: pick the layer, ▾ Edit or Generate, keep the result from ⏳ Pending - never leaving the full-screen canvas; the colour palette is Alt+P.

Settings ▸ About: the version, Produced and Developed by Tim Paul, Programmer - Claude Opus 5.5, and the credits to every project whose code, models, tools and ideas are used (links open in your browser); the same credits close the Guide.

Looking through the picked camera, ✥ Move / ⟳ Rotate show its arrows or rings in front of the view; the camera and the view move / turn together; Esc deselects it.

Anywhere in the app, typing in a box too: the prompt docker opens, unfolds, comes out of a closed ▣ docker and over a floating window or a maximized docker, the prompt box focused.

A picked camera lights up (no box) and shows X / Y / Z move arrows: drag the camera to move it at its height or an arrow to move it along one axis (1 cm steps); double-click it for the rotate rings and back; a click on another object picks that one.

A wall, floor, ceiling, skirting, cornice or floor slab clicked in the view or the outline shows its own panel: a wall's length, height and inside colour, the floor's level, the ceiling's type; Delete turns a part off.

30 built-in cross-sections in picture dropdowns; skirting swept with mitred corners (stopping at door frames), cornices under the ceiling, architraves round doors (both faces) and windows, window sills; ✎ Profiles… draws your own (points on a 1 mm grid, scale by width / height) and keeps them in the project.

Double-click a wall's top for a cut, drag it up / down (a gable end); pick a top edge and drag it or type its heights (a shed wall); Delete takes a cut / an edge away.

A room's ceiling None / Flat / Follows the walls' slope; a roof that follows them with an eaves width.

The docker and its 3D view use the theme's colours; a double-click on nothing deselects everything and orbits.

⛶ fills the whole bitGen window between the title bar and the status bar (then ⇲ Dock puts it back, 2.624.0), ✕ closes it; ☰ File: new, open, save, save as, rename, import, export, send, delete, close; 🖼 → Image editor and 🎞 → Timeline buttons; ⌂ Cut-away.

A bitGen tab in Blender's sidebar: send a scene or selection as glTF (or as a 3D layer), a live link that re-sends on every save, and get the project's 3D models into Blender.

Renders a Blender layout through its camera and has LTX 2.3 with 3DREAL make it photoreal.

Open a .blend over the timeline canvas: look through its camera or a free view, select, move, key, duplicate, hide and delete objects.

Set an In and Out, save only in bitGen, back to the original, or to a new file.

Render the segment and queue an LTX 2.3 3DREAL job.

Add a project 3D model to the open scene.

Turn a figure's bones; Blender keys the pose on save and render.

The scene as a 3D layer with this pose and view.

Render the animation to real, render a frame to a picture, send the scene as 3D to open or merge, or a figure as a 3D model.

Pick the figures, lights and cameras to merge into the open scene, the scene camera and the frame range.

Send a scene and open it as a 3D layer in the image editor.

BitGen moves the camera and poses in Daz Studio itself.

A Python-only plugin for the Unreal Editor that lists the project's levels, Level Sequences, cameras, MetaHumans and animations, answering only on this computer.

The port, an "alpha is inverted" switch, Test connection and how to install the plugin.

A level seen through one of its cameras, or the level viewport as it is now, as a picture layer.

A MetaHuman in the pose of an animation at a time, framed face, bust or full body, turned, on a clear background.

Nudge the camera's turn, tilt, dolly and field of view (or the figure's turn, pose time and framing), use the viewport now, render again.

A Level Sequence, or a level through a camera for a length you pick, rendered by the Movie Render Queue at the timeline's size and frame rate.

A MetaHuman playing an animation on a clear background, over the other layers.

Your own camera with keys of time, place, turn and field of view taken from Unreal's viewport or cameras, rendered through the level.

Each clip remembers how it was made and renders anew after the Unreal project or the keys change.

Move the camera of an Unreal or MetaHuman layer in the scene: drag to orbit, Ctrl to look, Shift to pan, Alt to roll, the wheel to dolly, Ctrl+wheel to zoom; small previews while dragging.

Lens presets (18-200 mm, macro, anamorphic, your own), focal length linked to the field of view, filmback, f-stop and focus distance (depth of field, click to focus), exposure (EV), place and roll / pitch / yaw, to and from Unreal's viewport.

Focal length, f-stop, focus and exposure for an Unreal camera's render.

Point, spot, rect (softbox), sun and sky lights with colour or temperature, intensity, reach, cone, size, volumetric scattering and shadows; presets (three-point, golden hour, night street, neon, softbox).

Exponential height fog (density, falloff, colour, start) with volumetric fog (albedo, extinction, distance) and fog presets.

Each light as a marker on the picture: drag it across, Shift-drag nearer or further, Alt-drag its aim.

The same lights and fog for an Unreal, MetaHuman or Unreal camera clip, or the image editor's ones brought over.

The timeline into one movie in the project folder.

MP4 H.264, H.265, H.265 10-bit, H.264 10-bit and AV1; MOV ProRes 422 HQ, ProRes 4444 with alpha and DNxHR HQ; WebM VP9; GIF; WAV and MP3.

Size, reframe (crop, pad or blur), frame rate and quality per format; packs add several at once.

A vertical or square version whose crop follows the subject.

HDR10, HLG or Rec.2020 on the 10-bit presets.

NVENC, Quick Sync, AMF or VAAPI when one works here.

Export just the ruler selection, or the frame at the marker as a PNG.

-14, -16, -23 or -24 LUFS.

Burned in, or an .srt beside each movie.

Clips, trims, layers, speed, moves, keyframes, opacity, volume, captions, markers and the grade as .cube LUTs; out and in.

For Final Cut Pro X, both ways.

CMX3600 EDL (V1) and OTIO, with markers.

The cut for Avid Media Composer and Pro Tools.

An XML or EDL opens as a new named timeline, clips matched by name.

The image editor's project, a zip of XML and PNGs.

A timeline with every layer, clip, keyframe, effect, caption, marker and its nested timelines, as XML.

The whole project, self-contained or as a manifest.

DXF and SVG sketches; STL, glTF and STEP bodies.

VR videos carry spherical metadata and the players' name suffixes.

The step-by-step instructions plus a one-page view of every shortcut key, in the studio's dark theme.

A 3-page brochure.

This catalogue of every feature by area (what it does, where it is, since which version), made again for every release.

Help pages for the timeline and the image editor.

The main window's own words (buttons, menus, docker titles, tabs, Settings, tooltips, hints) in Simplified Chinese, Traditional Chinese or Japanese, switched at once from a dropdown on the menu bar; the same setting as the guide's language, and the guides name the buttons as the app shows them. Settings is translated whole, including the status lines it writes while it runs (2.706.0), and every docker - the image editor, timeline, Sketcher, SolidCraft, StoryBoard and the rest - with their tools, panels and menus (2.707.0); the guides name the editors' buttons as the app shows them.

The 📖 Help window, Ctrl+K's guide entries, the ❔ help pages, the shortcut list and the Guide PDF in Simplified Chinese (mainland China), Traditional Chinese (Taiwan / Hong Kong, 2.704.0) or Japanese; the app's own buttons stay English, so the translations name them in English. A section newer in English says so, with the English one a click away.

Every language besides English is a language pack: ⤒ install one from a .zip (checked first; a language already there asks before it is replaced), ✕ remove it, see how much each has translated (the app, the guide, help, shortcuts, features); ⤓ Master pack - everything in English with TRANSLATING.txt - to translate bitGen Studio into a new language; ⤓ Export for translating gives a pack back with the newer English added, its translations kept.

Shared by the community lists the packs other people translated (from the bitGen Community on tigerart.com.au, each checked by an admin): ⤓ Install in one click; ⇪ Share sends an installed pack to the community as you, with a note for the admins; Your shared packs shows whether each is waiting, shared or not accepted (and why), ✕ takes one off. The website's Language packs page lists them too, with the master pack and an upload form.

The scene cards' ✎ First frame defaults to Auto: Qwen Image Edit 2511 in 4 steps when installed (Apache 2.0 - commercial use is fine), Qwen Image 2.1 edit for more than three pictures; a new Qwen Image 2.1 edit Turbo8 style runs Qwen 2.1 in 8 steps instead of 25 (Qwen Research License - non-commercial).

Every model file a render will load (the workflow's and the style's, Basic Settings' too) is checked against a list of the model files found in ComfyUI's model paths (misc/models_found.json, made again when a name is missing); a box lists the missing ones with the folder each goes in, ⬇ Download fetches the ones in the model catalog and the render then starts, Cancel stops it; a render that still stops on a missing model asks to download it (2.720.0).

Renders in a row that use the same model keep it on the card (no unload, no reload); another model, or an out-of-memory error, frees the card first; or free it before every generation as before.

💬 Community, Basic Settings, LLM tokens, the ▣ dockers, your own named dockers, floated main dockers, the stage docker and the voice chat share one order over the rest of the app: click one, or open it, and it comes to the top (a named docker no longer covers Community).

▶ on a song or sound in the preview starts a visualization picked at random (bars, mirror, waveform, ring, stars, ripples; 🎲 another one); Next / Prev go to the next / previous song of the docker it was clicked in and play it.

A midnight-navy theme with an electric-blue accent; the start-up splash is the intro film or the brain picture, at random; a new install's first project starts with the picture in Assets ▸ Images.

The prompt docker, Basic Settings, LLM tokens, the negative prompt docker and Tools and Properties.

Pause and play; shuttle backwards, stop and forwards.

The timeline's Select, Ripple, Roll, Slip and Slide tools.

Split, marker, In and Out.

Curve tools, shape tools, dimensions, the node Shape tool, the Outline Pen, Convert to Curves.

The sketch tools and their letters.

Screenshot and screen recording.