bitGen Studioby Tiger Art

Hardware guide

The editors run on any computer. The AI runs on your graphics card, and how much of it you can use depends mostly on one number: the card's memory (VRAM).

Don't have the hardware? bitGen Studio can render on a GPU you rent from RunPod by the hour, and run its language model on OpenRouter - see the RunPod and OpenRouter tutorials.

What each card can do

These are the rules the setup window uses when it checks your computer and recommends a package. The memory each model needs is an estimate; a card can do more with the right settings and less with other programs open.

NVIDIA card memoryNo graphics card8 GB12 GB16 GB24 GB or moreSystem RAM
Editing - every computer
Video and image editors, CAD and 3D, sound, export✓✓✓✓✓
AI helpers
Speech to text (transcripts, lyric timing)✓✓✓✓✓
Background removal✓✓✓✓✓
The assistant: screenplays, the prompt chat, ✨ Enhanceslow✓✓✓✓8 GB
Voices: text to speech and voice cloningslow✓✓✓✓
Splitting a song into vocals, drums, bass and the restslow✓✓✓✓
Pictures
Small models: SDXL, Anima–✓✓✓✓16 GB
Mid-size models: Krea 2 (GGUF), FLUX 3 Klein––✓✓✓16 GB
Big models: Krea 2 (full), FLUX.2 Klein 9B–––~✓32 GB
AI picture editing: Kontext, Qwen Image 2.1––~✓✓32 GB
Qwen Image Edit––––✓32 GB
A picture to a 3D model––~✓✓32 GB
Video
MiniMax H3: scenes, references, first and last frames–––~✓32 GB
LTX 2.5––~✓✓32 GB
Character swap, Viggle-Animate––––✓64 GB
AI upscaling (SeedVR2)–~~✓✓16 GB
DLSS 5 enhance and frame generation (NVIDIA RTX)–✓✓✓✓
Music
Songs: MiniMax Music 3, YuE2––~✓✓32 GB
  • ✓ runs well
  • ~ a smaller model or size, slower
  • slow runs on the processor, slowly
  • – not enough memory
  • MiniMax H3 needs 14 GB. On 14-20 GB it renders at 768 x 432 and splits long scenes into runs that fit; on 20 GB or more at 960 x 544. On a 16 GB card 16 GB mode turns on by itself.
  • A second graphics card is used for the assistant, which then stays loaded while the first card renders.
EditorRecommended without an NVIDIA card or with less than 6 GB.
Editor + small AIRecommended for 6-12 GB, and for a Mac with 24 GB of memory or more.
Editor + mid-size AIRecommended for 12-20 GB.
EverythingRecommended for 20 GB or more.

The packages are explained on the install page. Whatever you pick, you can add models later in Settings ▸ Models.

Memory, disk and processor

System RAM16 GB for pictures, 32 GB for video, songs and the big picture models, 64 GB for character swap. Models wait in system RAM while the card works, so too little makes everything load slowly.
DiskAn SSD with 100 GB free for the full set of models - the MiniMax H3 files alone are about 70 GB. The editor on its own needs a few GB.
ProcessorAny recent processor. It matters most without a graphics card, when speech, the assistant and voices run on it.
SystemWindows 10 or 11, Linux, macOS

NVIDIA, AMD, Intel and Apple

NVIDIAWhat the studio is built and tested on, and the only cards for MiniMax H3 and DLSS 5. Use a current driver: RTX 50-series cards need a CUDA 13 driver.
AMD / IntelComfyUI can run the picture models on them (AMD best on Linux with ROCm; Windows is experimental), but they are not tested with the studio, and MiniMax H3 and DLSS 5 need NVIDIA.
Apple SiliconThe editors run fully. The AI shares the Mac's memory - about 70 % of it can hold models - and small picture models, the assistant and voices run, slowly. The video and big picture models need an NVIDIA card.

Buying a card

  • Memory before speed. A model that does not fit is slow or does not run at all, while a slower card with enough memory just takes longer.
  • 8 GB is enough for pictures with the small models, not for video.
  • 12 GB adds the mid-size picture models, picture editing and songs, but not MiniMax H3 - a fast 12 GB card still cannot make its videos.
  • 16 GB is the sensible start for video: MiniMax H3 in 16 GB mode, LTX 2.5, songs and upscaling at full quality.
  • 24 GB or more runs the whole studio at full size.
  • Leave room in the power supply and the case: the big cards draw a lot of power and are long.
MemoryNVIDIA cards with it
8 GBRTX 5060, RTX 5060 Ti 8 GB, RTX 4060, RTX 4060 Ti 8 GB, RTX 3070, RTX 3060 Ti
12 GBRTX 5070, RTX 4070 / 4070 Super, RTX 4070 Ti, RTX 3080 Ti, RTX 3060 12 GB
16 GBRTX 5080, RTX 5070 Ti, RTX 5060 Ti 16 GB, RTX 4080 / 4080 Super, RTX 4070 Ti Super, RTX 4060 Ti 16 GB
24 GBRTX 4090, RTX 3090 / 3090 Ti
32 GB +RTX 5090 (32 GB), RTX 6000 Ada / RTX A6000 (48 GB), RTX PRO 6000 Blackwell (96 GB)

How fast is it?

Measured on an RTX 5090 (32 GB):

Picture edit, Qwen Image 2.1about 16 seconds
Picture edit, Qwen Image 2.1 Turbo8about 11 seconds
Video, MiniMax H3 Fast Omni at 540pa 10-second clip in about 1 min 12 s, a 15-second clip in about 1 min 48 s

More results, including an RTX 3090, are pinned in Discussions. Post yours there too - other people's cards help everyone choose.

No big card?

Start with the Editor package and add AI later, or rent a graphics card by the hour: Remote generation sends renders to RunPod and the assistant to OpenRouter, while everything else stays on your computer.

Check what you have

WindowsTask Manager ▸ Performance ▸ GPU: the card's name, and its memory as Dedicated GPU memory.
LinuxRun nvidia-smi in a terminal: the card and its memory.
macOSApple menu ▸ About This Mac: the chip and its memory.

You do not need to know beforehand: the setup window checks your card and memory, asks only if it cannot find them, and shows what each package will do on your computer before anything is installed. Get the installer.

Slide to download

0 / -

Slide to the end to start the download. It tells us a person downloaded it, not a robot - nothing about you is kept.