Summary
Our project received the official notice about the image quality incident affecting
Gemini 3 Pro Image (gemini-3-pro-image) and Gemini 3.1 Flash Image between
2026-06-18 2:22 PM PT and 2026-06-19 5:47 AM PT (blurry / low-detail results for
2K and 4K images, caused by a configuration error). The notice states the issue is
fully resolved. We would like confirmation, because we still observe suspected
quality degradation in a subset of generations produced after the incident window.
Environment
- API: Gemini Developer API (@google/genai SDK)
- Model: gemini-3-pro-image
- Config: responseModalities [“IMAGE”], imageSize “2K”
- Use case: technical line-drawing (blueprint-style) renderings of eyewear products
Observations
- Between 2026-06-23 and 2026-08-05 (i.e., after the announced incident window),
we collected 71 samples judged by a human reviewer as blurry / low-detail.
Sample images and generation timestamps (UTC) can be provided on request. - We also ran an objective sharpness analysis (variance of Laplacian) across
579 generations, comparing pre-incident vs post-incident outputs per view type:- Most view types show no significant difference between the two periods.
- However, our three-quarter-view generations show a ~19% drop in median
sharpness after the incident window (median 637 → 517; n=23 pre / n=63 post).
Questions
- Can you confirm there is no residual effect from the June 18-19 incident on
gemini-3-pro-image 2K generations after the announced resolution time? - Is the observed degradation on the affected subset a known issue or expected
model behavior?
No API errors are returned (normal successful responses) — only image quality is affected.