A ComfyUI instance, an API workflow, and its exact model/node dependencies.
A saved image, the workflow that produced it, and the exact prompt and seed needed to repeat it.
Inspect links before execution, then save the resulting image with workflow, seed, and model versions. Offline inspection does not prove execution.
A ComfyUI workflow that ran once and cannot be repeated is hard to build on. This skill guides your coding agent from prerequisites to a checked result: run one known workflow, diagnose the first failing step, and keep a recipe you can rerun.
Your agent inspects your exported API workflow and the local server's node definitions, lists missing nodes and models, and keeps the original graph and seed as a baseline. It changes one input at a time, shows the proposed inputs before queuing one image, and saves the output with its prompt ID. A job still running is collected by that ID rather than submitted again.
You keep the API workflow, prompt, seed, model source and license, and an expected-result note together for the next image. Identical pixels across hardware and versions are not promised.
You need a running ComfyUI instance, a compatible model and its companion files, and an agent with terminal access; start with the image models guide. Provider accounts, model weights, GPU compute and hosting are not included.
No reviews yet.
Sign in to rate this Vibe.
A form sends a prompt, each user gets back their own result, and every job shows clear states while it runs and when it fails.