ComfyUI Local Text-to-Image

Problem

After ComfyUI runs locally, the friction is often not model quality but repeated checks before every generation: whether the service is up, whether the port is reachable, and whether model files are in the right directories. This skill turns local text-to-image generation into a clear workflow: start from a text prompt, run Z-Image Turbo to create an image, and return the result in the conversation.

Workflow

It first checks that ComfyUI is reachable at http://127.0.0.1:8000 or 8188, and prompts for startup if the service is not responding. It then verifies that required model files are ready under diffusion_models/, text_encoders/, and vae/. The generation step can include a --negative argument for negative prompts. By default, it writes output_YYYYMMDD_HHMMSS.png at 1024×1024, uses a random seed, and delivers the image with a MEDIA: prefix.

Boundaries

This skill assumes an existing local ComfyUI installation and the specified models. It is not a remote image service, and it does not build or manage full ComfyUI workflows. If the port is unresponsive, models are missing, Python dependencies such as requests or pillow are absent, or generation times out after 60 seconds, the issue should be debugged in the local environment.