Generate Image

This node answers the question "can I make an image with the image generation I already configured in Open WebUI, then process it with ComfyUI's tools?" Yes - that's the whole job. ImageGenerate sends your prompt to your Open WebUI instance, which has an image-generation backend configured on its side, and drops the resulting picture back into the graph as a perfectly normal IMAGE tensor. ComfyUI never runs a diffusion model for this. The model that drew the picture lives behind Open WebUI - which is the point and the limitation at the same time.

Why you'd reach for it. You've got image generation working in Open WebUI, pointed at whatever the web UI can drive - a local engine, or a hosted/closed model you could never download. Now you want that result inside a ComfyUI workflow so you can do local things to it: upscale it, feed it through an img2img pass, mask or composite it, put a LoRA on it. That's the genuinely useful framing - use the Open WebUI-side model as a source of images, then treat the output like any other image in the graph. If your model is a hosted one, this is your door in, same as any API-wrapper node. If it's a model you actually own, you'd be better off running it locally in ComfyUI and keeping full control.

Be clear-eyed about what it is not. The README calls it "a basic version," and that's accurate. The only controls are connection and prompt - no size, no seed, no negative prompt, no steps. You get one image: the node takes the first item from the API response and ignores the rest. It's a convenience, not a replacement for a sampler.

How it works. Give it:

• connection (*) - from the Connection Node, carrying your IP, port, and API token.
• prompt (STRING, default "A cool car.") - what you want drawn.

It POSTs to http://ip:port/api/v1/images/generations with the prompt and your Bearer token, expects a JSON array whose first entry carries a url, fetches that image over HTTP, decodes it to RGB, and converts it to a torch tensor in 0–1 with a batch dimension. Outputs:

• connection (*) - passed straight through, so you can keep chaining other nodes that need it.
• image (IMAGE) - ready for Save Image, an upscaler, img2img, or anything that eats a batch of images.

Install, shared with the whole pack: ComfyUI Manager → search "Open WebUI API", or

cd ComfyUI/custom_nodes
git clone https://github.com/Gladon4/comfy-openwebuiapi

Restart ComfyUI and you're done. The only dependencies - pillow, requests, torch, numpy - are already in ComfyUI, so there's no extra install step.

Troubleshooting, in order of likelihood:

• Image generation isn't configured in Open WebUI at all. The README gates the whole node on it: "If you have set up image generation in Open WebUI." Verify the web UI's own image generation works in the browser first. If it doesn't, no node will fix it.
• Quality or content policy surprises. Both belong to the backend, not the node. If Open WebUI points at a hosted/closed model, its moderation is baked in - the node can't bypass a filter any more than a local one can.
• Host/port mismatch. You're calling Open WebUI's API directly, so the same rule as the Connection Node applies: use the address you actually browse.

If you already live in Open WebUI and just want its images flowing through ComfyUI's post-processing, this is a tidy little bridge. Just don't expect it to be a sampler.