Install bitGen Studio

Before you start

• Update your graphics driver first.
• Pick a big, fast drive - an SSD if you have one. The app, ComfyUI and the models all go in one folder there unless you choose others.
• Plan for an hour or two. Almost all of it is downloading: ComfyUI with PyTorch is several GB, the MiniMax H3 models about 70 GB.
• Optional: a free Hugging Face account and a read token, for the gated models - FLUX.2 Klein 9B and Lightricks' LTX files. Accept their licences on Hugging Face first.

Run the installer

Download the file for your system from the Download page and put it in an empty folder. It installs bitGen Studio into a Bitgen-Studio folder beside it.

The installer colours its lines: finished steps green, the step running now a lighter green, warnings orange, errors red. When a step is long and quiet, it says it is still working - it has not stalled.

The setup window

It opens the first time the studio starts and walks through these pages.

1. Your hardwareIt checks your graphics card, its memory (VRAM) and your RAM, and says what the studio can do with them.
2. Your packagePick how much of the studio to install. The one that suits your card is already picked, and you can add the rest later in Settings ▸ Models.

   Editor	No AI. The video and image editor, CAD and 3D sketching, sound and delivery. No ComfyUI, no models - ready in minutes on any computer.
   Editor + small AI	A 6-12 GB card. ComfyUI and the small models: Anima and SDXL pictures, background removal, speech to text, voices, a small assistant.
   Editor + mid-size AI	A 12-20 GB card. Adds Krea 2 and FLUX 3 Klein pictures, Kontext and Qwen 2.1 editing, MiniMax H3 video at a smaller size, songs and upscaling.
   Everything	A 20 GB+ card. The whole studio at full size: MiniMax H3 scenes, Qwen Image Edit, the big image models, LTX 2.5 and character swap.

3. ComfyUIInstall a new ComfyUI in a folder you choose, or use one you already have. A new one gets its own Python and the right PyTorch for your card - CUDA 13.0 or 12.8 for RTX 50, 12.8 or 12.6 for RTX 20-40, Metal on a Mac. Every custom node pack the studio uses is installed with it, then ComfyUI checks that each one loads.
4. Also set upOmniVoice (voice cloning and text to speech, about 3 GB), the assistant's engine llama.cpp (screenplays, the prompt chat and ✨ Enhance) and faster-whisper (transcripts and lyric timing). Viggle (native) is optional and unticked: about 5 GB plus the model parts your ComfyUI does not have yet.
5. InstallingEach step with its log. Settings that suit your card are made too: render size and length for your VRAM, where the language model runs, speech on the graphics card or the processor.
6. ModelsThe model files your package needs, ticked, with their sizes - see the models below.
7. ReadyWith nothing to report, the studio opens by itself a few seconds later and ComfyUI starts with it - you can make something right away. Anything that needs you is listed instead, and in setup.log in the app folder.

The models

Models come from Hugging Face. The setup lists every file your package uses, with its size, and the ones for your card are ticked.

• Without a Hugging Face token only the open models download. With one, so do the gated models whose licence your account has accepted.
• Download with the Hugging Face CLI (hf download - faster on big files) or the built-in downloader. Either one picks up where it stopped if the connection drops: run the setup again.
• Skip - I'll add models myself leaves them out. Settings ▸ Models ▸ ⤓ Model catalog is the same list inside the studio, any time.
• Before every render the studio checks the files it needs are there and offers to download any that are missing.

The MiniMax H3 video models go into ComfyUI's models folder:

Already have ComfyUI?

Pick Use a ComfyUI I already have in the setup window and browse to its folder - a portable build, a git install or a folder with its own Python environment. The setup shows which of the studio's node packs it already has and adds the rest. Your models stay where they are: bitGen reads ComfyUI's model folders and extra_model_paths.yaml.

Set it later in Settings ▸ ComfyUI: the folder (the one with main.py) and the address, http://127.0.0.1:8188 by default. More on the page for ComfyUI users.

No big graphics card?

The Editor package runs on any computer. For the AI parts, Settings ▸ Remote GPU renders on another machine's card:

• Another computer's ComfyUI - its address, for example http://192.168.1.20:8188, with ComfyUI started with --listen.
• RunPod - your API key lists your pods with their card and price; start one and use its ComfyUI.
• An SSH tunnel to a machine you can sign in to.

✓ Test asks that ComfyUI for its card and version, and Use it points the studio at it. The models a job uses must be on that machine.

Install by hand

Prefer to set things up yourself? Install these first:

Then get the studio and start it:

git clone https://github.com/Tim-byte-dev/Bitgen-Studio.git Bitgen-Studio
cd Bitgen-Studio
start.bat            # Windows
sh install_all.sh    # Linux / macOS

start.bat installs what is missing the first time and opens the studio every time after. On Linux and macOS, install_all.sh inside the folder uses that folder. No Git? Download the zip and unzip it - ComfyUI still needs Git.

Without the setup window

node scripts/setup-cli.js --comfy new --models recommended --hf-token hf_...
node scripts/setup-cli.js --detect          # the hardware check
node scripts/setup-cli.js --list-models     # the model ids
node scripts/setup-cli.js --only-models --models ltx23_omninft --downloader hf

Switches

Updating

Settings ▸ General ▸ ⟳ Updates: 🔍 Check for updates lists what is new, ⤓ Update now puts the newest version in place and ⟳ Restart now finishes it. Your projects, settings, styles, models and ComfyUI are not touched. The studio also checks quietly a little after it starts and tells you once when there is a new version.

From a terminal: git pull in the app folder, then start it again. What is new in each version is in the changelog.

Uninstalling

1. Close bitGen StudioStop ComfyUI too if it is still running.
2. Keep what you madeCopy the projects folder out of the app folder (and styles, graphs and custom.settings if you want your styles and settings).
3. Delete the app folderThe Bitgen-Studio folder holds the studio and everything it installed - and ComfyUI and the models, unless you put them somewhere else.
4. Remove the launcherLinux: ~/.local/share/applications/bitgen-studio.desktop and its copy on the desktop. macOS: ~/Applications/bitGen Studio.app.

Git, Node.js and Python stay installed - other programs may use them. Remove them the usual way for your system if nothing needs them.

When something goes wrong

Still stuck? Report a bug with the Console docker's output and the failed job's log from the Queue docker, or ask in the Discussions.