Introducing ChatGPT Images 2.5

Details

OpenAI has released ChatGPT Images 2.5, updating the image generation capability built directly into ChatGPT. The company describes the upgrade as producing more personalized, polished outputs that better reflect user intent when starting from ideas, sketches, or reference photos. The reference-photo framing suggests meaningful improvements to image-to-image workflows — a mode where users supply existing visuals as a style or composition anchor rather than generating purely from text. This would put ChatGPT Images more directly in competition with Midjourney's reference-image features and similar controls in Stable Diffusion. However, the official announcement is notably sparse: no model architecture details, no before/after comparisons, no benchmark results, and no information about what changed under the hood. It is unclear whether Images 2.5 represents a new underlying model or tuning applied to an existing one. For practitioners integrating OpenAI image generation into products or creative pipelines, the upgrade is worth testing — but the lack of technical disclosure makes it difficult to predict output quality improvements or whether it warrants a workflow change.