OpenAI's ChatGPT Images 2.5 Cuts Generation Time by 50% and Fixes Edits

OpenAI released ChatGPT Images 2.5, a substantial upgrade to its native image generation stack targeting the three most common complaints about Images 2.0: slow generation, subjects morphing between edits, and edits that quietly rewrite parts of the image you didn't touch. The release also adds two new API models, an in-chat drawing tool called Sketch, and template scaffolding for common formats like posters and merch.

Faster, more faithful generation

OpenAI cut image generation latency by up to 50% compared with Images 2.0, a meaningful improvement given how compute-heavy autoregressive image generation has been across the GPT-Image line. Beyond speed, Images 2.5 produces more natural lighting and richer textures, preserves subjects in reference photos more reliably, and follows editing instructions more consistently across multiple turns. That last point is the more interesting technical claim: in long ChatGPT threads, earlier edits are supposed to persist rather than degrade as you keep iterating, which has historically been a weak spot for both diffusion and autoregressive image models.

Region-scoped edits

Images 2.5 is better at editing only what you asked for while leaving the rest of the image untouched, even with complex subjects and backgrounds. You can now place a comment directly on a region to scope an edit, which is closer to how designers work in Figma than how people typically prompt image models. For API users, this means swapping out a single product, background, or piece of copy without rebuilding the entire asset, removing one of the main blockers for generative image models in production creative pipelines.

Sketch, templates, and shareable prompts

Three product-side additions ship with the release:

• Sketch: type @Sketch in ChatGPT to draw a rough layout, outfit contour, or doodle that the model uses as a visual guide for the final render.
• Templates: pre-structured formats for posters, flyers, merch, and product photos, so you're not starting from a blank prompt.
• Prompt sharing: when you share an image, you can attach the prompt that produced it so someone else can run the same recipe on their own photos.

Flare vs. Sunburst: the API split

GPT-Image-2.5 Flare is the default choice for most applications, delivering higher-quality images than GPT-Image-2 at 50% lower latency. It targets social content, product experiences, visual search, and high-volume generation. GPT-Image-2.5 Sunburst is built for premium visual workflows that need tighter control across edits, aimed at production campaign creative and polished product imagery, with longer generation times as the tradeoff.

Model	Best for	Tradeoff
Flare	High-volume, social, prototyping	Default speed and quality
Sunburst	Campaign creative, product shots	Slower, more precise edits

Early partner results

Higgsfield AI's product lead highlighted the edit-scoping behavior, noting the model understands what to leave unchanged while preserving character and composition. Manus reports Flare delivers images at two to four times the speed of GPT-Image-2 in their evaluations, with improved transparent-background generation, which matters for anyone generating UI assets or logos programmatically. Adobe confirmed the 2.5 models are coming to Firefly.

Availability and safety

Images 2.5 is available to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web, including free-tier accounts. API pricing is on OpenAI's business pricing page. On provenance, OpenAI continues using C2PA metadata and invisible watermarking to identify images made with its tools, and a system card covers the safety evaluations.

The shift to iterative editing

OpenAI reports users create more than 3 billion images per week across ChatGPT Images and the GPT-Image API, so latency and edit-consistency improvements compound quickly at that scale. The competitive pressure from Google's Gemini image models and a growing field of open-weights image editors has accelerated that push. More broadly, image generation is shifting from single-shot prompting to iterative, comment-driven editing, and Images 2.5 is among the first mainstream models built explicitly around that workflow rather than retrofitting it.