OpenAI image generator's output format is unpredictable, users prefer the previews

No_Body_4834 · reddit · 2026-09-12

A Reddit user complains that OpenAI's image generator returns unpredictable output formats: sometimes one image, sometimes a two-choice pair (likely for training), sometimes four alternatives, and occasionally (2% of the time) a "Preview" image before post-processing — which the user often prefers and has to save before it's replaced. Prompting for previews does nothing; asking for four images is often ignored or collapsed into a single composite image. The chat system claims it has no control over the image generator, and the user asks for any tricks to steer output format.

Original post →

More from Models

Models channel →