Card #410 · OpenAI Platform & APIs
gpt-image-1-mini generates nonsensical images unrelated to prompts
The gpt-image-1-mini model frequently produces images that completely ignore the prompt and input images, often generating random objects like food or humans instead. This issue is particularly prevalent when using the /edits endpoint or providing more than 4–5 input images.
Pain score
0.18/ 1.00 · weighted product of four components
Reach0.32distinct authors (log-normalized)
Recency0.99freshness decay, half-life months
Engagement1.00upvotes + comments, normalized
Monetization1.00willingness-to-pay cues in evidence
§ Evidence — 4 verified quotes
the generated result depicted a human instead of the animal
Other attempts with a similar prompt created images of a sandwich, cookies or a pizza, although no food was mentioned in the prompt or was visible in the uploaded images.
usually portraits that have nothing to do with your request, sometimes symbols or other images, but never anything related to what you actually requested.
Calls to /edits with several input images return bad images on about 95% of results.
Comments are a Pro feature — join the conversation, share what you've shipped, hear what others are building.
Upgrade to Pro