In plain English
- What happened
- On September 8, 2026, OpenAI released ChatGPT Images 2.5, which it describes as bringing "sharper details, more precise editing, and faster generation." It is "rolling out today to ChatGPT, ChatGPT Work, and Codex users across all tiers", which includes free accounts. The update adds a Sketch tool for drawing a rough layout, templates for formats like flyers, posters and product photos, and the ability to comment directly on an image to direct edits.
- Why it matters to your business
- Most small businesses make their own flyers, social graphics and product shots, and most of that work is redoing the same image with one thing changed. Precise editing that leaves the rest of the picture alone, plus templates for exactly those formats, is aimed at that job. And it is free, so the cost of finding out whether it is good enough for you is an afternoon.
- Do this
- Take the next flyer or product image you were going to make anyway and build it in ChatGPT using the Poster or product template, starting from a real photo of your product. Judge it on one question: would you print it? If yes, you have a new tool. If no, you spent thirty minutes and nothing.
OpenAI says people create "more than 3 billion images" a week across ChatGPT and its image API. Images 2.5 is the model behind that, and the release notes read less like a research announcement than a feature list for someone with a small business and no designer.
What is new
Three product features arrived with the model. Sketch lets you draw directly in ChatGPT and use the drawing as a guide; OpenAI's example is sketching the layout of a room, then adding a description of the style you want. You reach it by typing @Sketch in ChatGPT. Templates give you a starting point for "some of the most popular image formats, like flyers and product photos," with named options such as "Poster" and "Merch" where you add the information, design elements and style. And you can now place comments directly on an image to say what to change, instead of describing it in a fresh prompt.
Under the features, OpenAI claims three model improvements. It is "better at preserving the subjects in your reference photos," so a product or a person you upload should still look like themselves after the model moves them to a new setting. It is "better at editing only what you've asked for, while keeping the rest of the details the same." And generation is "up to 50%" faster than Images 2.0.
Available on every ChatGPT tier, including free, from September 8, 2026. The API versions, GPT-Image-2.5 Flare and Sunburst, are priced separately.
Try it on a real flyer
- 1
Start from your own photo, not a description
Upload a real picture of the product, the storefront, or the dish. The whole point of this release is that reference photos survive the edit. A described product is a generic product.
- 2
Pick the template and fill in the facts
Choose Poster or the product-photo template and give it the real details: the offer, the date, the price, the phone number. Templates are there so you are not starting from a blank prompt.
- 3
Edit by commenting, not by starting over
If the price is wrong or the background is busy, comment on that part of the image. The model is built to change only what you point at. Re-prompting from scratch throws away the parts that were already right.
- 4
Check the text with your own eyes
Read every word and number on the finished image before it goes anywhere. Image models have historically been weakest at text, and OpenAI's claim is that content is "more accurate," not that it is always right.
The two things to be careful about
First, the model's strength with reference photos cuts both ways. A picture of your product in a new setting is a marketing asset. A picture of a customer or an employee in a setting they were never in is something you need their permission for, and putting a recognizable person into a generated scene without asking is the fastest way to turn a free tool into a complaint.
Second, everything it makes is labelled. OpenAI says it continues to use "C2PA metadata and invisible watermarking to help identify images made with our tools." That is not a problem for a flyer. It is worth knowing before you use a generated product photo anywhere that implies it is a photograph of the actual item, because the file itself says otherwise.
Reality check
A generated product photo is an illustration of your product, not a photograph of it. Use it for the poster, the social post, the menu board. Do not use it where a customer will reasonably assume they are seeing the exact thing they will receive. The model improved; the rule did not change.
The honest read is that this is a version update, not a new category. What makes it worth a look is where OpenAI pointed it: at flyers, merch, product photos and the person who makes them at 9 p.m. If that is you, the templates are the part to try first.
Sources
