[Question] How to use koboldcpp comfyui in open-webui ?

[0fflineuser]

Hi,

I am trying to setup open-webui to use koboldcpp for image generation and edition.

Image generation works properly with automatic1111.
I use these settings in open-webui for image generation and it works:

But open-webui only supports openai/comfyui/gemini for image edition, but I can't figure out how to set it up. In the koboldcpp readme it is written that the comfyui api is supported but I didn't find any explanation on the wiki.

Here are the settings for image edition in open-webui:

What should I set them to? How do I find the nodes ids ?

edit:
Here are the results of the tool calls in open-webui.
For image generation:

It sometimes fails on the first request but then works. Maybe while the model is loading as I use autoswap.

Here for image edition:

For image edition it keep retrying and it never works.

[henk717]

We do not emulate comfyui to my knowledge but the UI of ours can connect to comfyui, but you could try it with the emulation for dalle.

[0fflineuser]

We do not emulate comfyui to my knowledge but the UI of ours can connect to comfyui, but you could try it with the emulation for dalle.

Yes, it seems that the openai endpoint works. But now I get this error:

It seems that kobold fails to load previously generated images.
Also I am not sure if it's intended behavior but the images I generate from open-webui don't show up in the https://<koboldcppinstance>/sdui under the Images tab.

I also asked(from open-webui) to edit an image that was generated from the koboldcpp stableui directly, just to see if it works (I assume it must work if we know the image id, but maybe I am wrong):

By the way the img2img works properly when done from the koboldcpp stable ui directly, so it's not an issue where it already doesn't work.

[henk717]

Previous images or content is not a thing on KoboldCpp. The UI's are completely standalone web pages, as a server it doesn't store history. You would need image history on the openwebui side or manually transfer in between them.

[0fflineuser]

Ok.
By the way I kept automatic1111 for the image generation as the openai one doesn't work for image generation, so maybe it also doesn't work for image edits ?

[LostRuins]

The OpenAI image endpoint is at http://localhost:5001/v1/images/generations , is that correctly configured?

There is no disk storage - previously generated images are not saved on the server and cannot be "retrieved" later except for the most recent image (at some times)

[0fflineuser]

@LostRuins It seems the openai endpoint doesn't work.

Here is a video:

screen_2026-08-07.07.42.43.mp4

edit:
The issue seems to be with the model loading. I tried the /v1/images/generation again after the /sdapi/v1/txt2img one ran and it worked.

[0fflineuser]

Also, is there a plan to add the openai /v1/images/edits endpoint for image editing ?
Since we agree right now there is no way for it to work with open-webui, or did I misunderstood something ?

[LostRuins]

@0fflineuser good idea, i've added it and it will be in the next release, also the issue should be fixed. New rolling builds are now compiling