OpenAI rolled out ChatGPT Images 2.5 on September 8, 2026, and it’s the biggest change to ChatGPT’s picture-making tools since they launched. You can now sketch a rough drawing and have ChatGPT turn it into a finished image, edit one part of a picture without regenerating the whole thing, and share a prompt so a friend can remix it with their own photo. If you’ve ever given up on an AI image because you couldn’t quite describe what you wanted, this update is built for exactly that problem.
Here’s the 10-second version of what changed:
- Released September 8, 2026, to every ChatGPT plan, plus ChatGPT Work and Codex
- Sketch is the headline feature — draw directly in the chat and ChatGPT builds a full image from your lines
- Generation is up to 50% faster than the previous version, Images 2.0
- New in-image commenting and Templates make small fixes and common formats (posters, logos, product shots) much quicker
- Developers get two new API models: GPT-Image-2.5 Flare and Sunburst
What’s Actually New in ChatGPT Images 2.5
The short version is that ChatGPT’s image generator got sharper, faster, and easier to correct. OpenAI says generation latency dropped by roughly 50% compared to Images 2.0, so the annoying wait between a prompt and a usable result is much shorter now.
Reference-photo handling also improved. If you upload a photo of yourself, a product, or a pet and ask ChatGPT to put it in a new scene, the update does a better job keeping faces, logos, and proportions consistent across edits. Lighting and texture also look more natural, less like the slightly plastic sheen older AI images had.
None of this is a totally new model in the GPT-6 sense — it’s an upgrade to the image pipeline that sits alongside chat models like GPT-6 Astra. Think of it as OpenAI polishing a tool you already use rather than launching something from scratch.
How the Sketch Feature Actually Works
Sketch is exactly what it sounds like: a drawing canvas inside ChatGPT. You rough out shapes, arrows, or a basic layout with your mouse or finger, add a short text prompt describing the style you want, and ChatGPT fills in the details.
It’s aimed at a specific kind of frustration — when you know exactly what you want but can’t find the words for it. A lopsided circle labeled “logo here, blue and gold” communicates more than three paragraphs of prompt text ever could. Early testers have found it especially useful for quick mockups, room layouts, and logo placement ideas.
The feature works on desktop, mobile, and web, so you can sketch on your phone during a commute and finish the image later on a laptop.
Templates, Comments, and Shareable Prompts
Sketch gets most of the attention, but three smaller additions matter just as much day to day. Templates give you a guided starting point for common formats like posters, merch mockups, and product photos, instead of staring at a blank prompt box.
In-image commenting lets you click directly on the part of a picture you don’t like — a stray shadow, an odd hand, a wrong color — and describe the fix right there, rather than rewriting your entire prompt and hoping the rest of the image doesn’t change. Shareable prompts let you send an image along with the exact instructions that made it, so someone else can drop in their own photo and get a matching result.
Together, these turn ChatGPT’s AI image generator from a one-shot novelty into something closer to a real editing tool.
GPT-Image-2.5 Flare vs. Sunburst: What Developers Get
Alongside the consumer update, OpenAI shipped two new API models for developers building their own apps on top of GPT-Image-2.5. They cost the same per token — the difference is speed versus precision.
| Model | Best For | Speed | Editing Control |
|---|---|---|---|
| GPT-Image-2.5 Flare | High-volume, general-purpose apps | Fastest — the default option | Standard |
| GPT-Image-2.5 Sunburst | Production creative work, brand assets | Slower, prioritizes quality | Tighter, more precise |
On pricing, OpenAI charges the same rate for both — around $8 per million image input tokens and $30 per million image output tokens — so the choice comes down to whether a project needs raw speed or fine-grained control, not budget.
Does It Beat Google’s Nano Banana 2?
Google’s Gemini image tool, nicknamed Nano Banana 2, has been the benchmark AI image generators get compared against all year. Reviewers who tested ChatGPT Images 2.5 in its first day found the speed gap has closed noticeably, and image quality is now close enough that the “winner” often depends on the specific prompt rather than one model being clearly better.
Where ChatGPT still pulls ahead is workflow — Sketch, Templates, and in-image commenting don’t have a direct match in Gemini’s toolset yet. If you’re already living inside ChatGPT for chat and work tasks (or using the Gemini app for Windows alongside it), this update makes it less likely you’ll need to open a second app just for image editing.
Frequently Asked Questions
What is ChatGPT Images 2.5?
It’s OpenAI’s September 2026 update to ChatGPT’s built-in image generator, adding a Sketch drawing tool, Templates, in-image commenting, shareable prompts, and roughly 50% faster generation.
How do I use the Sketch feature in ChatGPT?
Open the image tool in ChatGPT, choose Sketch, draw a rough layout with your mouse, finger, or stylus, add a short text description, and ChatGPT generates a finished image based on your drawing.
Is ChatGPT Images 2.5 free to use?
Yes, it’s available to every ChatGPT plan, including the free tier, though paid plans generally get higher usage limits and faster access during busy periods.
What’s the difference between GPT-Image-2.5 Flare and Sunburst?
Flare is the faster, general-purpose option and the API default. Sunburst is slower but gives developers tighter editing control, aimed at production-level creative work rather than quick, high-volume generation.
Does ChatGPT Images 2.5 work on mobile?
Yes. Sketch, Templates, and the rest of the update work across desktop, mobile, and the ChatGPT web app, so you can start an image on your phone and finish editing it later on a computer.
ChatGPT Images 2.5 is one of those updates that sounds minor on paper but changes how you actually use the tool — Sketch alone will save a lot of people from typing three paragraphs to describe a drawing they could have made in ten seconds. It’s not a reason to cancel a Gemini subscription if you already like Nano Banana 2, but if you’re an existing ChatGPT user who’s avoided the image tool because it felt fiddly to correct, this is the update that fixes that.
Leave a Reply