SAN FRANCISCO — ChatGPT users can now upload a photo of themselves and see a generated image of a jacket, dress or pair of glasses on their own face and body, as OpenAI pushes the chatbot further into online shopping.
The company rolled out the tool globally on Thursday, alongside a Favorites feature that saves products inside the app. A “Try on” button appears on clothing and accessory listings in ChatGPT shopping results on mobile and the web. Tap it, take or upload a selfie, and add a full-body photo if asked. ChatGPT Images then produces a preview. Users can also drop in a screenshot or a photo of an item and ask the chatbot to try it on.
OpenAI’s help documentation is explicit about the limit. Try-on images “may not represent the product or your appearance exactly and do not guarantee fit or size.” The company tells users to check the merchant’s measurements, product details and return policy before buying. The preview is a picture, not a fitting room.
The image model underneath is ChatGPT Images 2.5, which OpenAI introduced on Sept. 8 and describes as better at keeping people recognizable from reference photos, with more natural lighting, richer textures and more reliable editing. Lower latency is part of the pitch. A try-on that takes too long, or that changes the shopper’s face, fails the only test that matters.
Reference photos are saved so the same selfie can be reused. They can be replaced or deleted under Settings, then Personalization, then Reference photos. Saved try-on images sit in the Library next to favorited products. OpenAI’s help pages do not describe extra privacy controls specific to try-on photos beyond that delete option.
Favorites is the quieter half of the update. Products found in a chat can be stored and sorted into folders, so a search for a coat does not disappear when the conversation does. The same update adds a multi-page document scanner in the iOS camera, a separate tool that has nothing to do with clothes but ships in the same release.
Shopping in ChatGPT already surfaced product cards with images, prices and links when a question looked like a purchase. Ads are labeled separately from those results. The new layer is visual and personal. A user can describe a style and ask for the pieces, upload a celebrity outfit and ask for items on sale that resemble it, or start from a listing and generate a version worn by the account holder.
Google has offered a similar try-on in shopping for more than a year, which is why the feature looked familiar the day it launched. Amazon and a string of fashion retailers have run their own versions, with the same complaint: the drape is approximate, the size is not measured, and hair or background sometimes changes without being asked. The Verge noted that in one OpenAI example the tool restyled the subject’s hair. That is the tradeoff of a generative preview. It is not a photograph of the garment on the body.
For retailers, the appeal is fewer returns and a reason to stay inside the chat instead of opening a brand site. OpenAI does not fulfill the order. The listing still points to a merchant. What the company gains is another reason to open ChatGPT before a browser, and another category of user image stored on its side of the account.
The privacy question is the practical one. A try-on requires a photo of the shopper’s face and, for clothes, the body. Those images persist until the user deletes them. They are used to generate further try-ons. OpenAI has not, in the material released with the feature, said how long a deleted photo remains in backups, whether try-on outputs are used to train models, or how a minor’s photo would be handled if one were uploaded. The control the company does advertise is the settings toggle.
Fit is the other gap. A generated image cannot know whether a size medium matches a shoulder or a waist. OpenAI says so in the help text. Shoppers who treat the picture as a measurement will learn that at the return counter, which is the outcome the feature is supposed to reduce.
The launch is another consumer use for a chatbot that began as a text box. Images, shopping cards, saved products and now a virtual garment on a user’s own photo are the sequence. The button is live. The caveat is in the documentation: the picture is not a promise of how the clothes will look, or how they will fit.
