What is ChatGPT Images 2.5 and how does the Sketch tool work?

TL;DR. ChatGPT Images 2.5 launched September 8, 2026 and is rolling out to all ChatGPT tiers. It generates images up to 50% faster than Images 2.0, produces sharper results with more natural lighting, and introduces Sketch — a drawing canvas you invoke with @Sketch that lets you draw a visual reference instead of describing one in words. Multi-turn editing is also more precise: it now changes only what you ask and leaves the rest of the image intact.

---

ChatGPT's image generation got a substantial upgrade today, and the part most worth knowing about isn't the quality bump — it's a new way to give the model visual instructions that doesn't require you to be good at writing prompts.

What's new in ChatGPT Images 2.5

OpenAI describes Images 2.5 as the company's "state-of-the-art image model," released five months after Images 2.0. Four changes stand out for professionals:

Faster generation. Images 2.5 reduces generation latency by up to 50% compared to Images 2.0. In practice this means you spend less time watching the progress bar and can iterate faster — useful when you're refining something across multiple edits.

Better output quality. The model produces more natural lighting, richer textures, and more realistic detail — particularly on portraits, product shots, and anything with fine surface texture like fabric, wood, or food. It also handles complex layouts and transparent backgrounds more reliably, which matters for anything you plan to composite into other designs.

More faithful subject preservation. When you upload a reference photo and ask for a variation, Images 2.5 does a better job holding the subject's features consistent across generations. If you're creating multiple images with the same person, product, or object, this reduces the drift that made earlier versions frustrating for anything requiring consistency.

Precision editing across multiple turns. The editing behavior changed in a meaningful way: Images 2.5 now modifies only the element you asked to change and leaves the rest of the image as-is. With Images 2.0, asking to "make the background sunset orange" would sometimes shift the foreground colors too. The new model treats edits as surgical by default.

Sketch: the part that changes how you prompt

The most interesting new feature is Sketch — an in-app drawing canvas that you invoke by typing @Sketch in a ChatGPT message.

Here's the problem it solves: describing visual ideas in text is genuinely hard. "Put the logo in the top right, the headline below it centered, and a product photo taking up the left two-thirds" is a reasonable art direction note, but getting ChatGPT to interpret it correctly — especially for layout-heavy work — often requires multiple back-and-forth iterations. Sketch shortcuts that: you draw the rough layout yourself, and ChatGPT uses your drawing as a visual reference rather than guessing from text.

What you can do with Sketch:

• Draw a rough composition (boxes, shapes, stick figures are fine — accuracy isn't the point)
• Add text labels to indicate what goes where
• Choose from built-in templates for common formats: "Poster" and "Merch" are currently listed, with more likely to follow
• Combine the drawing with a text prompt to fill in what your sketch doesn't specify

You can also place comments directly on a generated image to indicate specific areas for editing — rather than saying "the thing in the bottom left," you mark it.

Sketch is designed for people who think visually but aren't graphic designers. It's particularly useful for rough ideation (do I want this layout before I describe it in detail?) and for cases where spatial positioning matters more than artistic quality.

Who gets Images 2.5

Per OpenAI's September 8, 2026 announcement, Images 2.5 is rolling out to all ChatGPT, ChatGPT Work, and Codex users across all tiers — Free through Enterprise — on desktop, mobile, and web. The announcement doesn't specify tier-specific restrictions or different image allowances by plan.

Usage limits still apply per plan — OpenAI doesn't publish exact image counts, but Free users have a lower quota than paid tiers. Your live remaining allowance shows under Settings → Limits in ChatGPT.

The new API models: Flare and Sunburst

For developers and teams using the API, OpenAI released two new image models alongside the ChatGPT update:

• GPT-Image-2.5 Flare — the standard API model. Same quality and speed improvements as the ChatGPT version, now the default choice for most applications. OpenAI says it delivers higher-quality images than GPT-Image-2 at 50% lower latency.
• GPT-Image-2.5 Sunburst — the precision-focused variant. Longer generation time, tighter editing control, aimed at workflows where accuracy matters more than speed: product photography, detailed illustration, precision composition.

API pricing for both models: $8.00 per 1M image input tokens, $30.00 per 1M image output tokens, with cached-input rates available.

What this means for professionals using ChatGPT for visual work

The speed improvement alone changes daily use — iterating on visuals across a conversation feels more like real-time design feedback and less like a slow batch process. The editing precision improvement is significant for anyone doing multi-step refinement; you won't need to re-describe the entire scene every time you want to change one element.

Sketch is the bigger shift in the longer run. It addresses the core weakness of text-to-image prompting — the mismatch between how people think about spatial layouts and how well those layouts translate into words. For professionals who use ChatGPT for marketing assets, presentation visuals, social content, or rough concept mockups, Sketch lowers the barrier to getting something usable faster.

Whether the output quality now clears the bar for client-facing work depends heavily on the use case. ChatGPT's image generation has consistently been stronger on photorealistic images than on clean design layouts — that likely holds with Images 2.5. For polished graphic design work, it's still a starting point rather than a finished product. For content-adjacent visuals — blog header images, social tiles, mood boards, concept sketches — the faster iteration and better quality makes this version materially more useful than Images 2.0.

---

Sources

• OpenAI — "Introducing ChatGPT Images 2.5," September 8, 2026: https://openai.com/index/introducing-chatgpt-images-2-5
• The Verge — "ChatGPT Sketch turns your bad drawings into detailed AI images," September 8, 2026: https://www.theverge.com/ai-artificial-intelligence/991727/openai-chatgpt-images-2-5-sketch
• 9to5Mac — "OpenAI releases ChatGPT Images 2.5 with 'sharper details' and 'more precise editing'," September 8, 2026: https://9to5mac.com/2026/09/08/openai-releases-chatgpt-images-2-5-with-sharper-details-and-more-precise-editing/
• Unite.AI — "OpenAI Releases ChatGPT Images 2.5 With Sketch and Two New API Models," September 8, 2026: https://www.unite.ai/openai-releases-chatgpt-images-2-5-with-sketch-and-two-new-api-models/