Skip to content

Getting started

The tool has three parts that run on your PC: the plugin inside Unreal® Engine 5.8, a small local studio server, and a local language model. This page gets all three running. The MVP is in development, so details may change between builds.

What you need

  • Windows 10 or 11 with one NVIDIA GeForce RTX graphics card. More VRAM means a larger model next to the editor; 16 GB is a comfortable start, 24–32 GB leaves room for the model and the scene.
  • Unreal Engine 5.8.
  • Python 3.11 or newer.
  • LM Studio for the local model (Ollama works too).
  • ComfyUI (Comfy Desktop) for frames and video, optional; the studio finds it at http://127.0.0.1:8188.

Mac and Linux are not supported yet. Without an NVIDIA graphics card, generation runs in the cloud: with your own provider keys on the free plan, or with the plan's credit on a paid plan (see pricing).

1. Install the plugin

  1. Download the latest early-access build from My account › Downloads (see the changelog for what is in it).
  2. Unzip it into the Plugins folder of your Unreal project, so you get YourProject/Plugins/<plugin folder>.
  3. Open the project. Enable the plugin and the built-in Python Editor Script Plugin if Unreal asks, then restart the editor.

2. Start the local model in LM Studio

  1. Download a 7–14B model that keeps JSON well, for example Qwen3 14B in Q4_K_M.
  2. Open the Developer tab, load the model and press Start Server. LM Studio listens on http://127.0.0.1:1234.
  3. Set the context length to at least 8192 tokens.

3. Start the studio server

In the server folder that came with the build:

pip install -r requirements.txt
uvicorn app:app --port 8765

The server listens on http://127.0.0.1:8765. It uses port 8765 because Unreal 5.8's built-in MCP server already takes 8000. If your model is not in LM Studio, set LLM_URL to its OpenAI-compatible endpoint, for example Ollama at http://127.0.0.1:11434/v1/chat/completions.

4. First scene

Write a short dialogue in the plugin's director panel and generate. The server sends it to the model, checks the scene JSON against the schema (asking the model to fix mistakes up to twice) and queues it. The plugin then builds the Level Sequence: characters on marks, cameras, voice lines and cuts.

5. From scene to film

Frames and video are made by ComfyUI on your PC through the Vishot app (the studio server), or in the cloud with your own provider keys or a paid plan's credit. Install Comfy Desktop and start it once: the studio finds it at http://127.0.0.1:8188 and skips any step whose models are missing. You can install models on your PC from My account › Models. If your card cannot hold the video model, switch video to the cloud.

Next: connect the plugin to the portal to sync model sets between machines.