Skip to content

Install

Castcut is the app; ComfyUI does the rendering and an LLM writes plans and checks pictures. You need all three — Castcut can install the LLM for you.

What you need

  • A GPU with 24 GB for the full set of tools (12–16 GB runs stills and short clips).
  • ComfyUI on this machine or your network.
  • About 100 GB of disk for the common models, more for every video tool.

1. Get Castcut

Download from GitHub Releases:

System File
macOS .dmg
Windows .exe
Linux .deb (best) or .AppImage

The first launch opens Settings → ComfyUI and runs Heal & ready for you. More in Desktop app.

git clone https://github.com/doodersrage/castcut.git
cd castcut/docker
cp .env.example .env      # ComfyUI, the LLM, a sign-in password + secret
docker compose up -d --build
docker compose logs -f castcut

Open http://<host>:47832. ComfyUI on the Docker host is http://host.docker.internal:8188 (start it with --listen). Optional extras:

  • --profile llm — an LLM in a container (NVIDIA GPU) — the built-in LLM can't run inside the Castcut image; see Built-in LLM in Docker
  • --profile comfyui — ComfyUI in a container

More in Configuration.

Needs Node.js 22+.

git clone https://github.com/doodersrage/castcut.git
cd castcut
npm install
cp .env.example .env.local   # set COMFYUI_API_URL
npm run dev

Open http://localhost:47832.

2. Connect ComfyUI

  1. Settings → ComfyUI → Connection — enter the address (http://127.0.0.1:8188 on the same machine).
  2. Settings → Overview → Heal & ready — installs the node packs Castcut uses and turns on its workflows.
  3. Settings → ComfyUI → Models — install the models for what you want to make. See Models.

Tip

The Castcut node pack (Settings → ComfyUI → Castcut nodes) makes checks and cut-outs run inside one ComfyUI job. Optional, but faster.

3. Connect an LLM

Pick one:

Option When
Built-in LLM (Settings → LLM) Easiest. Installs llama.cpp and one vision model into Castcut's data folder; starts when needed, stops when idle. In Docker, run it as a container instead: Built-in LLM in Docker.
LM Studio or Ollama You already run one. Castcut finds it at the usual address.
A hosted provider Settings → LLM → provider and key, per browser.

A vision model matters: it reads stills for the checks (faces, outfits, poses, adult age).

4. Check it

Settings → Overview → Play checks lists every check and the model or node pack it needs. Green means ready. Then go to Your first film.