API Generator support

[jalfort]

Would it be possible to add support for API based generation like FAL.ai, Replicate and OpenAI, basically providing an API key so you could generate images straight into Krita and then use the AI tools to enhance and potentially resend pictures back to Fal.ai or Replicate etc.

Would be a cool concept, Invoke has some interesting support on the paid plan but not in the community version.

[tarelion]

Hi, I’m also interested in this direction.

I think there may be a useful middle ground between the current local/ComfyUI workflow and fully separate external tools like ChatGPT, Krea, FAL, Replicate, etc.

The local ComfyUI backend is great for fast iterative inpainting/outpainting and canvas-aware edits, but there are cases where artists may want to send the current canvas, selection, or reference image to a stronger external model for higher-level concept exploration, typography-heavy generation, or instruction-based edits, then bring the result back into Krita as a layer.

Would the project be open to a generic “external image provider” interface, rather than adding one vendor-specific integration?

For example, a provider capability interface could describe support for:

• text-to-image
• image-to-image / reference image
• selection or mask-based edit
• outpainting
• async job status / progress
• result import as a new Krita layer

Then specific providers could be implemented separately, e.g.:

• OpenAI API
• FAL / Replicate
• custom HTTP endpoint
• community-maintained bridges

For my own workflow, I’m especially interested in using Krita as the canvas UI while routing some jobs to a stronger external model, while keeping local ComfyUI as the default path for fast edits.

I’m asking before attempting any implementation because I want to understand whether this direction fits the project scope. If it is in scope, I’d be interested in helping prototype a small provider interface or a custom HTTP provider first, instead of starting with a large vendor-specific PR.