OpenAI launched a virtual try-on inside ChatGPT on 1 October 2026: the user uploads a selfie and a full-body photo, and the model renders the garment on them. The feature runs on September's ChatGPT Images 2.5 model. It ships alongside a favourites library, and photos from personal accounts can end up in training.
Key takeaways
- Input is a selfie plus a full-body photo, output is a generated preview of the garment
- Works with any clothing image, including a screenshot from outside the ChatGPT catalogue
- The engine is ChatGPT Images 2.5, released in September 2026
- A new library saves products and generated try-ons for later
- Photos uploaded from a personal account may be used for training unless the user opts out
What actually changes
The try-on lives inside a product card in the conversation but is not tied to it — ChatGPT also accepts a screenshot from any store. The model can additionally identify a garment worn by someone in an uploaded photo and look for available equivalents.
On the model side the change is incremental. Images 2.5 is said to deliver more natural lighting, richer textures and faster generation — the only differences claimed, with no benchmarks or figures attached.
What the user actually hands over
A portrait is not enough for the feature to work. It needs a full-body photo — material that allows body proportions to be reconstructed, not just a face. Combined with purchase history, that yields a profile far sharper than any number of text prompts. The default consent setting leaves that decision to the user, who has to find it in account settings.
| Item | Status |
|---|---|
| Input | selfie + full-body photo, optional product screenshot |
| Model | ChatGPT Images 2.5, September 2026 |
| Library | saves products and generated try-ons |
| Countries and subscription tiers | not disclosed |
| Comparative data for Images 2.5 | not published |
Virtual try-on is not new
Google shipped a comparable feature earlier, so OpenAI is not opening the category so much as joining it with its own image model. The difference is the entry point: rivals start a try-on from a shopping search, here it starts from an ordinary conversation.
The business context is less rosy. The earlier Instant Checkout feature underperformed, and the company has begun testing ads inside chats. The try-on fits the same attempt to convert conversational traffic into transactions.
Consent here follows an opt-out?Opt-out: A consent model where the feature is on by default and has to be switched off by the user — unlike opt-in, where it must be switched on first. model, and the setting itself is not part of the try-on flow.
Why it matters
A corpus of full-body photos is far more sensitive than a typical text prompt, and training consent being on by default shifts the burden onto the user. The commercial value is obvious — a try-on shortens the path from browsing to buying and keeps it inside a single chat window.
What's next
- OpenAI has not said which countries and subscription tiers get the feature
- Consent to use photos for training has to be switched off manually in account settings
- The company has published no comparative data for Images 2.5 against the previous version
