bitGen Studio

A local, self-hosted AI film and design studio. Write a screenplay with a local LLM, render scenes with MiniMax H3, LTX and Wan through ComfyUI, cut them on a full timeline, paint and compose in a layered image editor, sketch and model in CAD (constrained sketches and parametric 3D bodies), and bring in scenes from Blender, Daz Studio and Unreal Engine 5.8. All of it runs on your own GPU, with no accounts and no cloud, and nothing leaves your machine.

Warning

Public beta — still in testing. bitGen Studio is a beta and is still being tested. Expect rough edges and the odd bug. Bug reports and improvements are very welcome: open an issue (the Bug report and Feature request templates are there to help), or send a pull request. See docs/BETA.md for how to report a problem and CONTRIBUTING.md for how to help.

Where testing is up to

Area	Status
🧩 Model manager	✅ Tested by the developer on real hardware - works
🖼️ Image generation (every built-in style)	✅ Tested on real hardware - works
🎬 Video generation	✅ Tested on real hardware - works
🎵 Music	✅ Tested on real hardware - works
💬 Local LLM (chat, screenplays, ✨ Enhance)	✅ Tested on real hardware - works
📦 Installing from this repository (Windows, the setup wizard: ComfyUI, PyTorch, the node packs)	✅ Tested on a fresh install - works
🖌️ The editors (image editor, timeline, CAD sketches and 3D bodies)	⚠️ Built and covered by the automated tests, but not yet tested enough with real projects
🏠 House builder	⚠️ Built and covered by the automated tests, but not yet tested enough with real projects
🎮 Unreal Engine and Daz Studio bridges	🧪 Experimental - tested only against stand-in servers, not inside the real programs

If you try one of the ⚠️ or 🧪 areas, your reports are especially useful.

Install · User guide · Feature reference (PDF) · Guide + shortcut keys (PDF) · Changelog · Technical reference · MIT licence

---

What it does

The Feature Reference lists every feature (700+), what it does, where to find it and the version it arrived in. In short:

🎬 Films and music videos	A local LLM writes a screenplay, either a film whose characters speak in their own cloned voices or a music video cut to a song's segments. The scene board renders each scene with MiniMax H3 (video and sound), LTX 2.3 / 2.5 or Wan 2.1 / 2.2. You keep, redo, upscale and produce one film from the kept takes.
🧑‍🎤 Characters and swaps	Character sheets, voices, clothing and identity models that follow a character through every scene. Character swap replaces the person in any video, VR footage included, and keeps the performance and the sound.
🎞️ Timeline	A preview canvas with layers, keyframes, effects and adjustment layers. Also: colour grading, masks and tracking, multicam, nesting, proxies, captions, a track mixer, transitions, motion-graphics templates and real-time playback with a render bar. AI tools work inside the timeline, and you can export to FCP7 XML, FCPXML and AAF.
🖌️ Image editor	Layers, masks, smart objects with smart filters, and adjustment and fill layers. Selections, painting and retouching, warps (puppet, liquify, perspective) and type with OpenType features. It also has paths and vector shapes, drawing tools, frames, artboards, dimensions, channels, actions (.atn too), documents, PSD and PDF, and ✨ AI generate / fill / edit.
📐 CAD	Sketch layers with real constraints and dimensions: trim, extend, fillet, offset, patterns, DXF / SVG in and out, and measure. A parametric 3D body layer supports extrude, revolve, fillet, chamfer, shell and split body, with named parameters. It is built by OpenCascade in the browser or by Blender, and exports STL, glTF and STEP. A CAD ribbon runs along the top of the image editor and the timeline.
🎮 Unreal Engine 5.8	A plugin links the Unreal Editor. You get Unreal layers and MetaHuman layers in the image editor, and Unreal, MetaHuman and Unreal camera layers on the timeline (rendered by the Movie Render Queue). You can move the camera (orbit, look, pan, dolly, roll) and set its lens, f-stop, focus and exposure. You can add lights, volumetric lights and volumetric fog, all from inside bitGen.
🧊 3D, Blender and Daz	3D layers on the timeline (models, 3D text, materials, lights, bakes), a Blender add-on and render bake, Daz Studio scenes as 3D layers with a live camera and pose link, and image-to-3D.
🖼️ Images	Named image styles (FLUX.2, SDXL, Z-Image, Qwen-Image…), LoRA stacks, ControlNet, face fix and SeedVR2 upscaling.
🎵 Music and sound	Songs (MiniMax Music3, YuE2), stems, voice cloning and TTS, speech-to-text, karaoke with a bouncing ball, and SFX.
🖥️ Yours	A frameless desktop app (or any browser), a job queue, a gallery, a live console, dockable panels and themes. Projects stay on your disk.

Screenshots

The pictures, video and music in these screenshots are placeholders made for the page. The app was running without models, so nothing in them was generated.

The main window (Basic workspace). The References docker holds the pictures a prompt can use, each with a short description (a red sports coupe, a green velvet chesterfield). The prompt refers to them as <Picture 1>, <Picture 2> and so on. The preview plays the selected take, and the gallery keeps every render. The Workspace menu at the top switches between layouts: Basic, Image editing, Video editing, Music video and Expert.
🧩 Model manager. Every model and LoRA in your ComfyUI folders is shown as a card with its family (FLUX.2, SDXL, LTX, Wan, Qwen, Z-Image…) and kind. Search and filter them, make a new image style from a model, or download from Civitai into the right folder. Tested on real hardware.	🖼️ Image styles (Settings ▸ Image). A style is a model, its ComfyUI workflow, a sampler and a LoRA stack saved under one name. It ships with built-in styles for FLUX.2 / FLUX 3 Klein, SDXL, Pony, Illustrious, Qwen Image Edit, Krea 2, Anima, FLUX Kontext and more. Pick one beside the prompt. All built-in styles tested on real hardware.
🎞️ Timeline (Video editing workspace). Video, audio and image tracks, its own preview canvas, a tools and properties panel and a layers list. Clips can be trimmed, keyed, graded and given effects, then compiled to one film. Beta: needs more real-world testing.	🎵 Music and stems. Write lyrics and a style caption, pick an engine (MiniMax Music3, YuE2) and generate songs. Stems splits a song into voice, drums, bass and the rest. Tested on real hardware.
🖌️ Image editor (Image editing workspace, full screen). Layers, masks, adjustments, selections, paint and retouch tools, type, paths, CAD sketches and 3D layers, plus ✨ AI generate / fill / edit. Here a picture has a cut-out layer on top of it. Beta: needs more real-world testing.
🏠 House builder. Draw walls, rooms, doors and windows on a plan, furnish them from a library, see the house in 3D, then place cameras for shots and 360° locations that the video renders can use. Beta: needs more real-world testing.

Install with one file (Windows, Linux, macOS)

Download one file from this repository into the folder you want bitGen Studio in (a big, fast drive - ComfyUI and the models need about 100 GB) and run it. It installs Git and Node.js when they are missing (Windows: Python too), clones this repository next to it into Bitgen-Studio, and starts the studio - the first start installs everything else and opens the setup window. Run the same file any time later to open the studio.

System	File	How to run it
Windows 10 / 11	install_all.bat	double-click (say Yes when Windows asks to install Git / Node.js)
Linux	install_all.sh	sh install_all.sh (asks for your password once if Git has to be installed)
macOS	install_all.command	sh install_all.command in Terminal (or chmod +x it once, then double-click)

Install from GitHub (Windows)

1. Install Git - git-scm.com (or winget install Git.Git).
2. Install Node.js - nodejs.org, the LTS version (18 or newer).
3. Get bitGen Studio - open a terminal in the folder you want it in (a big, fast drive is best: ComfyUI and the models need about 100 GB) and run:

   git clone https://github.com/Tim-byte-dev/Bitgen-Studio
   

4. Run start.bat in the new Bitgen-Studio folder. Start installs bitGen Studio if it is not installed yet and runs it if it is - so from now on, start.bat is also how you open the studio. The first time it:
   • checks for Python (3.10 or newer). If it is not found, start says so: install it from python.org with Add python.exe to PATH ticked, then run start.bat again;
   • installs the Node and Python packages and the app window (about 150 MB, once), then opens the setup window.
5. Choose the version you want installed - the setup window shows what your computer can run and recommends one: Editor (no AI), Editor + small AI (6-12 GB graphics card), Editor + mid-size AI (12-20 GB) or Everything (20 GB+).
6. ComfyUI - if you already have ComfyUI, pick Use a ComfyUI I already have and select its folder; if you do not, the setup window installs ComfyUI for you. Either way it adds the node packs and downloads the models for the version you chose, then opens the studio.

Keep the black console window open while you use bitGen Studio - closing it stops the studio.

_{The Expert workspace. Basic Settings on the left, the References docker with a picture as <Picture 1>, the prompt chat turning "horse in picture 1 trotting" into a full shot, the prompt ready to Generate, and the queue and gallery with the finished renders on the right.}

docs/INSTALL.md has every step in detail, the one-file installers, the local LLM set-up and troubleshooting.

Quick start

1. Install: follow Install from GitHub above, or download the latest release from the Releases page. Unzip bitgen-studio-v<version>-full.zip and run start.bat (Windows) or misc/bitgen-studio.sh (Linux / macOS). You can also clone this repository and run the same script. Either way you need Node.js 18+, Python 3.10+ and Git. On first start it installs its Node packages and Python environment; Settings ▸ ComfyUI and the setup wizard bring in ComfyUI, its node packs and the models. To build the one-file installers yourself (they bring their own Node.js, Python, git and ffmpeg), use scripts/build-installers.sh. docs/INSTALL.md has every step, the LLM set-up and troubleshooting.
2. Make something: type a prompt and press Generate, or ask the assistant to create a screenplay and render the scenes on the board. The user guide (also as a PDF) walks through every docker.
3. Found a bug or have an idea? Please open an issue. This is a beta, and every report helps.
4. UI dragged off the screen, or a black window? Close bitGen Studio and run misc\reset-layout.bat (in the app folder). It puts back the default layout and theme, and clears the desktop app's caches. Your projects, settings and models are kept, and everything it resets is copied to backups\layout-<time>\ first.

Optional bridges (each in its own folder):

• blender/bitgen_bridge: the Blender add-on.
• daz/bitgen_bridge: the Daz Studio scripts.
• ue/bitGenBridge: the Unreal Engine 5.8 plugin. Copy it to <YourProject>/Plugins/.

Requirements

	Minimum / notes
OS	Windows 10/11 (developed on Windows 11) or Linux. macOS runs the studio, but MiniMax H3 video is built for NVIDIA CUDA.
GPU	NVIDIA with CUDA. Developed on an RTX 5090 (32 GB). 24 GB+ is the realistic floor for MiniMax H3 video; the image editor, CAD and timeline editing run on any GPU.
RAM / disk	32 GB RAM or more; ~100 GB free disk for ComfyUI, the models and your projects.
Software	Node.js 18+, Python 3.10+, Git and ffmpeg on the PATH. Optional: Blender 4.x (the Blender engine, bakes, add-on), Daz Studio, Unreal Engine 5.8.

How it fits together

bitGen Studio window (Electron, or http://127.0.0.1:8311 in a browser)
   │
   start.bat → misc/server.js ── routes/ + lib/   Node: the API, projects, the job queue (server.js stays minimal)
   │
   ├─ ext/<id>/          the dockers (timeline, image editor, video, music…) - each an extension
   ├─ static/            the page, shared modules, vendored libraries (three.js, pdf.js, OpenCascade…)
   ├─ misc/              server.js, the Python workers (*_worker.py - one per kind of job, each driving ComfyUI over
   │                     its API), default.settings, the setup scripts (get-*.ps1, reset-layout.bat …) and the notes
   ├─ llama.cpp          the local LLM: screenplay writer, chat, ✨ Enhance
   └─ bridges            Blender (blender/), Daz Studio (daz/), Unreal Engine 5.8 (ue/)

Tests

bash scripts/e2e-stub-test.sh runs the whole suite: 1,400+ legs of unit and real-browser tests. It runs against stub ComfyUI, LLM and Unreal servers, so no GPU or models are needed. GitHub Actions runs a quick subset on every push (scripts/ci-units.sh). See CONTRIBUTING.md.

Licence

MIT: see LICENSE and NOTICE.md. Generation software (ComfyUI, llama.cpp, the audio engines) runs as separate programs that the app downloads from their own projects.

The browser libraries in static/vendor/ keep their own licences, which sit beside them. One of them, OpenCascade (static/vendor/replicad/), is LGPL-2.1. NOTICE.md explains how it is shipped.

✨ Generate in the 📐 SolidCraft docker uses cadrille (Kolodiazhnyi et al., ICLR 2026 - paper, weights): an open 2B model that writes CadQuery code for a part from a description or a picture; bitGen Studio runs that code and rebuilds the part as SolidCraft's own editable sketches and extrudes. cadrille's code is Apache-2.0, but its weights are CC BY-NC 4.0 - non-commercial use only. They are not part of this repository: Settings ▸ Add-ons ▸ CAD generate downloads them from Hugging Face to your machine, and you accept their licence when you do. Without cadrille, the app's own local LLM writes the code instead. See NOTICE.md.

Model weights have their own licences. MiniMax H3, FLUX.2, SDXL, the voice models and others each come with their own terms, and several are non-commercial. Read the licence of the model that rendered your output.

Links: the software bitGen Studio bridges to

Software	Link	What bitGen Studio does with it
ComfyUI	github.com/Comfy-Org/ComfyUI	The generation engine: every image, video and music style runs as a ComfyUI graph. The app starts it, installs its node packs, reads its model folders and edits its extra_model_paths.yaml.
llama.cpp	github.com/ggml-org/llama.cpp	The local LLM (llama-server) behind ✨ Enhance, the prompt chat, screenplays and ✨ Generate's code.
Blender	blender.org	The bitGen Blender add-on: scenes and renders both ways, a .blend opened in the preview canvas, ✨ Build a scene with AI, and Send to Blender from SolidCraft.
Unreal Engine	unrealengine.com	The bitGen Bridge plugin: Unreal / MetaHuman layers in the image editor and timeline, renders from a camera, and SolidCraft parts imported as Static Meshes.
Daz Studio	daz3d.com	The bitGen Daz bridge: renders, pictures, scenes (open / merge) and figures.
SolidCraft	solidcraft.io (source)	Open-source parametric CAD in its own 📐 docker: designs in the project, pictures to the image editor, 3D layers to the timeline, parts to the House builder, Blender and Unreal.
cadrille + CadQuery	cadrille · CadQuery	✨ Generate in SolidCraft: a description or a picture becomes CadQuery code, rebuilt as editable SolidCraft features. cadrille's weights are CC BY-NC 4.0 (non-commercial).
Civitai	civitai.com	The 🧩 Model manager: details, trigger words, thumbnails, example pictures and videos for your models, plus search and download.
Hugging Face	huggingface.co	Where the model files come from (Settings ▸ Models, Add-ons, the Model manager's downloads).
Final Cut Pro, DaVinci Resolve and other editors	OpenTimelineIO · DaVinci Resolve · Final Cut Pro	The timeline out and in as FCP 7 XML, FCPXML, EDL and OTIO.
Krita AI Diffusion	github.com/Acly/krita-ai-diffusion	The image editor's selection Generate / Fill / Remove follows its technique per model family.
RunPod (or any remote GPU)	runpod.io	Settings ▸ Remote GPU: render on a ComfyUI on another machine or a cloud GPU.
FFmpeg	ffmpeg.org	Compiles, exports, proxies and the audio / video conversions.

Acknowledgements

ComfyUI · cadrille + CadQuery · SolidCraft · MiniMax H3 · LTX-Video · Wan · llama.cpp · stable-diffusion.cpp · three.js · replicad + OpenCascade · pdf.js · pdf-lib · ag-psd · fontkit · and the ComfyUI node packs listed in NOTICE.md.