gemini-3-pro-image checks the input before generating. When it blocks, it returns HTTP 200 within seconds with promptFeedback.blockReason: OTHER and no candidates. In a 6-image fashion-catalog request we tested, the only trigger was one AI-generated person reference image, independent of the prompt, and flash passed. After that image was re-exported as JPEG, it passed and the full request generated normally. The new FAQ covers how to find the image by splitting the set in half, plus preprocessing advice for manual and code workflows: Why Does Gemini Image Return blockReason: OTHER?
A prompt that asks for text: when the prompt contains sentences such as “produce a design proposal” or “analyze in depth,” image-only calls return NO_IMAGE some of the time. In our tests this happened about 21% of the time, with the same request failing and then succeeding. Appending one sentence, “output the final image only,” brought it to 0/24. This is now covered in NO_IMAGE troubleshooting.
← Back to Live Updates · 📚 Monthly archive