bitGen StudioTiger Art 出品

关于

bitGen Studio - a local AI video, image, music and editing studio. Everything runs on your own computer: no sign-ups, no cloud.

Tiger Art

Tiger Art 是澳大利亚的一家独立软件工作室。我们制作的是 bitGen Studio：一个在你面前这台电脑上运行的完整电影工作室，让作品、文件和决定权都留在你自己手中。

它免费开源，采用 MIT 许可证。每一行代码都在 GitHub 上——欢迎阅读、修改，并把改进提交回来。

制作者

  • Produced and Developed byTim Paul
  • ProgrammerClaude Opus 5.5

致谢

bitGen Studio stands on the work of many people. Thank you to everyone below - the projects whose code, models, ideas and tools we use or learned from.

Generation engines

  • ComfyUI by Comfy Org and comfyanonymous - the generation engine - every image, video and music style runs as a ComfyUI graph; its own frontend's graph conversion is mirrored for workflows
  • comfy-kitchen by Comfy Org - the int8 ConvRot weight layout, re-implemented in the native engine (Apache-2.0)
  • llama.cpp by ggml-org - the local LLM (llama-server) behind ✨ Enhance, the prompt chat, screenplays and the vision captioner
  • stable-diffusion.cpp by leejet - the first image and video backend (sd-cli)
  • Hugging Face diffusers by Hugging Face - the MiniMax H3 weight conversion and modular pipeline the native engine follows (Apache-2.0)
  • Hugging Face Hub by Hugging Face - where the model files come from
  • FFmpeg - compiling, exporting, proxies and every audio / video conversion

Video models

  • MiniMax H3 by MiniMax - the main video model - text, references, first / last frame, native audio
  • Viggle-Animate by Viggle AI - character animation; its inference code (Apache-2.0) guided the native engine
  • LTX-Video by Lightricks - the LTX video styles and the official templates
  • Wan by the Wan team - the Wan video styles, camera paths (ATI) and VR worlds
  • SeedVR2 by ByteDance Seed - upscaling takes
  • FastH3 Omni - the FastH3 recipe for MiniMax H3 graphs
  • Veda - the optional sparse-attention accelerator for MiniMax H3
  • MiniMax-H3-NativeAudio-MusicVideo-Workflow by Shrek3OnVH5 - the music-video lip-sync workflow the MMH3_NativeAudio_MV style is built from (MIT)

Image models

  • Qwen Image and Qwen Image Edit by the Qwen team, Alibaba - picture styles, ✏ Edit, fills and the Sketcher's real photo
  • FLUX.2 and FLUX 3 Klein by Black Forest Labs - picture styles, edits and LoRA training bases
  • Krea 2 by Krea - picture styles (Turbo and Raw)
  • FIBO 2.0 by Bria - the Fibo 2.0 Edit styles
  • Anima, MOSS-Image, SDXL and Pony models - more picture styles
  • SAM 2 (Segment Anything) by Meta - Select Subject, cut-outs through a shot
  • Hunyuan3D 2.1 by Tencent - 🧊 picture to 3D (under its own community licence)
  • TRELLIS.2 by Microsoft - 🧊 picture to 3D
  • cadrille by col14m - ✨ Generate in SolidCraft writes part code from a description or a picture (weights CC BY-NC 4.0)
  • CadQuery - runs that part code

Language, speech and music

  • Qwen3, Qwen3.5, Qwen3.8 and Qwen3-VL by the Qwen team, Alibaba - the recommended local LLMs, the vision captioner and fine-tune bases
  • Gemma 4 by Google - local LLMs and fine-tune bases
  • WeVisOne-VL - the vision add-on
  • faster-whisper by SYSTRAN - transcription, karaoke words, edit by transcript, voice chat
  • Whisper by OpenAI - the speech-recognition models faster-whisper runs
  • OmniVoice - text-to-speech and voice cloning
  • OpenAudio / fish-speech by Fish Audio - the first text-to-speech engine
  • Ming-3.0 and Bark - the speech add-ons
  • YuE - song generation (YuE2), as its own engine and in ComfyUI
  • Demucs by Meta - stem separation (vocals / instrumental)

Training

  • ai-toolkit by Ostris - the 🎓 LoRA and checkpoint trainer
  • Unsloth - LLM and vision fine-tunes

ComfyUI node packs

  • ComfyUI-KJNodes by Kijai - the MiniMax H3 long-video nodes and many helpers
  • ComfyUI-GGUF by city96 - GGUF model loading
  • ComfyUI-VideoHelperSuite by Kosinkadink - loading and saving video and audio in graphs
  • comfyui-inpaint-nodes by Acly - the image editor's selection fill
  • comfyui_controlnet_aux by Fannovel16 - ControlNet preprocessors
  • ComfyUI-MiniMaxH3Mod, Spectrum-MiniMax-H3, ComfyUI-H3-NativeAudioLock - character RefMods, MiniMax H3 speed-ups, the song lock for music videos
  • Nougan Nodes, RH DLSS 5, SolarRenderer - regional character LoRAs and Prompt Relay, ⚡ DLSS 5, ☀ solar rendering

Built into the app

  • three.js - every 3D view (MIT)
  • opentype.js - 3D text (MIT)
  • pdf.js by Mozilla - opening PDFs in the image editor (Apache-2.0)
  • pdf-lib and fontkit - saving PDFs with embedded fonts (MIT)
  • ag-psd by Agamnentzar - PSD files and their text effects (MIT)
  • polygon-clipping - the Sketcher's walls (MIT; with splaytree and robust-predicates)
  • replicad and OpenCascade - parametric 3D bodies (MIT; OpenCascade LGPL-2.1)
  • jxl-oxide, libjxl and @jsquash/jxl - JPEG XL open and save (MIT / Apache-2.0 / BSD-3-Clause)
  • Node.js, Express and Multer - the app's server
  • Electron - the desktop app
  • Python, NumPy, OpenCV, soundfile and pyaaf2 - the workers, tracking, beats, sound and AAF export
  • OpenTimelineIO - timeline exchange (OTIO, FCP XML, EDL)
  • uv by Astral - the setup's Python
  • ReportLab, pypdf, jsdom and Playwright - the Guide PDFs and the test suite

3D models

  • Female mannequin model by dimasalcalde (CGTrader) - the Sketcher's 🧍 mannequin (model) - People in the Furniture tool
  • Female mannequin model by zahers (CGTrader) - the Sketcher's 🧍 mannequin (model) - People in the Furniture tool

Text effects and fonts

Programs bitGen Studio works with

  • Blender by the Blender Foundation - the Blender add-on: scenes and renders both ways, 3D layers, poses
  • Unreal Engine and MetaHuman by Epic Games - the bitGen Bridge plugin: Unreal layers, cameras, lights, the live view
  • Daz Studio by Daz 3D - the Daz bridge and the live link
  • SolidCraft by Tangency Labs - open-source parametric CAD in its own docker
  • Civitai - model details, examples, search and download
  • RunPod - rendering on a remote GPU
  • DaVinci Resolve and Final Cut Pro - timelines out and back in

Ideas and inspiration

  • Krita AI Diffusion by Acly - the image editor's selection Generate / Fill / Remove follows its technique per model family
  • Wan2GP and mmgp by deepbeepmeep - the idea of streaming model blocks to the GPU (no code used)
  • The Sketcher's furniture library - the furniture, people and materials - first made for the House builder of earlier versions
  • The professional creative tools - the photo editors, vector and drawing programs, video editors, motion-graphics tools, 3D modellers and parametric CAD programs whose ways of working the image editor, the timeline, the CAD tools and the Sketcher follow
  • Everyone who tests bitGen Studio - and reports what to fix