Muse Studio: local video workflows with ComfyUI

Muse Studio's creator demonstrates a local video workspace with a Node.js interface and a Python backend using LangGraph. Setup covers dependencies, environment files and startup scripts. Cloud model endpoints are optional; the example uses Ollama with what the speaker identifies as Qwen 3.5 32B.

ComfyUI supplies media generation. The tutorial labels input and output nodes, exports API JSON and imports workflows into Muse Studio. The speaker recommends native Save Video after reporting output-handling problems with Video Helper Suite.

The example generates 24 story scenes and tracks prompts, keyframes and approval on a Kanban board. Z Image Turbo creates character sheets, though the speaker warns that results may need retries. A Flux 2 reference workflow creates scene keyframes, followed by an LTX 2.3 video example. Playground outputs require explicit assignment to a character or scene. The local version disables collaborator mode because of memory and runtime costs.

LangGraph agents assist editing, and Remotion renders an export that the speaker trims and renders again. These are basic editing controls; the speaker suggests later upscaling. Out-of-memory recovery is discussed, but the video generation succeeds in the narrated example. For export, the speaker switches projects because the earlier project's media files had been deleted.