Workflows

FlixML is a layer on top of ComfyUI. Every generation runs a workflow — a ComfyUI graph exported as JSON, with a few values (prompt, dimensions, seed, character) filled in at request time.

Built-in workflows

FlixML ships with workflows for image and video generation out of the box. List what’s available on your install:

curl <your-api-url>/api/workflows        # the catalog: what each one makes and needs
curl <your-api-url>/api/workflows/<id>   # one workflow in full, every param explained

Each workflow has an id you pass to the generate endpoints. The catalog leaves out each workflow’s params — they run long, and you only need them for the one you picked, which is what the second call is for.

Adding your own

1. Build and test a graph in ComfyUI.
2. Export it as API-format JSON.
3. Drop the JSON into the workflows directory and give it an ID.
4. Reference that ID when you generate.

FlixML injects request-time variables (prompt, width/height, seed, character trigger words, LoRA weights) into the graph before queuing it. Authoring details — which nodes get filled, how variables map — are in app/flixml/workflows/registry.py in the repo.

How a job flows

1. You call POST /api/image/generate (or /api/video/generate) with a workflow ID and parameters.
2. FlixML resolves the workflow, injects your variables, and picks a provider by role.
3. The job is queued on that GPU’s ComfyUI and tracked to completion.
4. The finished media is imported and organized automatically.

You get a prompt ID back immediately — poll it or browse results in the Studio UI.