Opened 2 months ago
Last modified 6 hours ago
#64918 new defect (bug)
WP 7.0 Connectors: No models found that support image_generation
| Reported by: |
|
Owned by: | |
|---|---|---|---|
| Milestone: | 7.0.1 | Priority: | normal |
| Severity: | normal | Version: | trunk |
| Component: | AI | Keywords: | ai-client |
| Focuses: | Cc: |
Description
Immediately after generating an image with AI, the modal offers the option to Refine.
However, I keep getting the error No models found that support image_generation for this prompt. after trying to refine the image that was just generated.
Steps to Reproduce
- Go to Settings > Connectors and add a valid API key for OpenAI
- Create a New Post and add an Image Block
- Click on "Generate Image" and type in a prompt
- After the image is created, click on
Refine Image - Type in a prompt for the image refinement
Expected
The image should be updated.
Actual
The site throws an error saying there are no models that support image generation.
Attachments (1)
Change History (9)
This ticket was mentioned in Slack in #core by mukeshpanchal27. View the logs.
8 weeks ago
#7
@
7 weeks ago
After investigating, I think this is more likely a plugin capability/declaration issue. With Gemini image refinement works fine. With OpenAI, the error shown is "No models found that support image_generation for this prompt," that suggests this is likely a plugin capability-mapping issue. The current error, “No models found that support image_generation for this prompt,” is probably happening because OpenAI provider still advertises GPT image models with text-only input.

Similar to #64917, we should ensure there's better direction for a user on how to resolve this issue. Perhaps something as noted in https://core.trac.wordpress.org/ticket/64917#comment:5:
Separately, this is something we've wrestled with in the AI plugin and whether we should disable feature/experiment availability if a corresponding provider+model-capability is not available. However if someone has this working as expected and then runs out of tokens/credits or there's an issue with a provider's API, then we run into the same problem. Curious for folks' input on most optimal way to handle for various scenarios here.