GPT-Image
OpenAI's image generator inside ChatGPT, running the November 2026 GPT-Image-2.6 model. The leader for both creative work and the editing that goes into UI design.
My take
Almost flawless. I used to find the older versions too generic and reached for Google's models instead; now this is my daily driver. What won me over first was the editing, being able to point at one part of an image and change just that. What keeps me here is how well it works out what I'm actually trying to visualize. I enjoy it for creative work as much as for the unglamorous editing that goes into UI design. Version 2.6 solidifies the lead: it feels faster and handles a long series of edits from one prompt flawlessly.
At a Glance
Pros
- + Built directly into ChatGPT for zero-friction conversation
- + Precise image editing allows clean modification of specific areas
- + Works out what you were trying to visualize from a rough description
- + In my use, text renders cleanly and legibly
- + Handles a long series of edits from one prompt, in my use
- + Makes animated GIFs and keeps reference images of people, pets and characters for reuse
- + About 50% faster than the previous version, according to OpenAI
- + Available in ChatGPT on every plan, including Free (with limits)
Cons
- − Less artistic stylization than Midjourney
- − High-volume generation can hit ChatGPT message caps
Best for: Image editing and precise modifications directly inside ChatGPT conversations, plus creative work and UI design layouts
Powered by
Full review
GPT-Image is OpenAI’s image model inside ChatGPT. The current version, GPT-Image-2.6, appears there as ChatGPT Images 2.6. It shipped on 4 November 2026, two months after 2.5, and is available in ChatGPT and Codex on every plan.
A tool for image editing
The defining feature is editing. Instead of re-generating a whole image when you want a small change, which used to scramble details, you can point at specific areas and describe the modification. The model edits the selected part and leaves the rest alone. That precision is what makes it usable in real work rather than as a toy.
The 2.5 release made this hold up over longer sessions. The failure mode of conversational image editing has usually been the fourth edit quietly undoing the second, and it survives a run of them much better now. With 2.6 I can put a whole series of edits into one prompt, and it works through them flawlessly.
What actually improved
With 2.5, not what you’d look for first. The version before it was already strong enough that side by side, the output doesn’t announce itself. The difference shows in comprehension: hand it a rough description of an idea and it’s much better at working out what you were trying to visualize, then picking a clever way to show it. That matters more than another notch of fidelity, because the slow part of making an image was never the rendering. It was the four attempts it took to explain what you wanted.
With 2.6 it’s hard to say how much plain generation and editing improved. It feels faster, and OpenAI says both API models generate about 50% faster than their 2.5 versions. The visible addition is animated GIFs, which are neat, even if they’re not the most important thing for business work. You can also keep reference images of people, pets and characters and reuse them in later images.
Typography and layouts
In my use, the text-rendering problems that plagued earlier image generators are gone, and 2.6 keeps it that way. Text comes out clean and legible, which is what makes it usable for structured layouts: posters, mockups, and diagrams that need readable text.
Two jobs, one tool
Most people end up with separate tools for making images and for fixing them. This covers both. I use it for creative work and for the unglamorous editing that goes into UI design, and it’s my first choice. For a second opinion I run the same request through Nano Banana. Midjourney still wins when you want a particular look rather than a particular idea.
Who should use GPT-Image
Anyone who needs to iterate on visuals, edit existing images, or design layouts with clean text. Its placement inside ChatGPT makes it convenient, and the editing makes it genuinely useful for day-to-day design work. The underlying model is covered in the model entry. In the API there are two variants: gpt-image-2.6-flare, the fast one, and gpt-image-2.6-sunburst, for detailed work that has to stay consistent across edits. ChatGPT doesn’t let you choose; Higgsfield does.
Added: 2026-02-21 · Last updated: 2026-11-07