ChatGPT Try On won't promise your size: what sellers publish

No. ChatGPT Try On shows a garment on a photo of you, and OpenAI is reported to say its images may not reproduce the person or the product exactly and do not guarantee size or fit. The feature launched on October 1 2026, so a shopper can now see a look before buying, but still has to trust your size chart for the number on the tag.

That split is the useful part for a seller. The picture does the "will I like it" job; your listing has to do the "will it fit" job. Below is what the launch coverage says, what that means for the assets you publish, and what Sume's try-on Formats can and cannot add.

What has OpenAI actually launched?

Per TechCrunch, OpenAI announced on Thursday October 1 2026 a global rollout of two shopping features: a Try On button in shopping results, and Favorites, which saves products to a Library next to your try-on images. A shopper uploads a selfie or full-body photo, or a screenshot of an item, and the feature is built on ChatGPT Images 2.5.

The fit caveat comes from Retail Technology Innovation Hub, which reports that reference photos are kept for future try-ons, can be changed or deleted in settings, and that OpenAI notes generated images may not match the person or product exactly and do not guarantee size or fit. We could not open OpenAI's own page (it returned 403 to our fetch), so treat both as reported, not as OpenAI's wording.

What does a try-on image not tell a shopper?

A generated image is a plausible picture, not a measurement. It can drape a medium shirt on a body that would need a large, and it cannot know that your fabric has no stretch. Google's own description of its earlier try-on says the model reflects how a garment would drape, fold and cling, and it shows garments on real models from XXS to 4XL (Google, page dated June 14 2023, read 2026-10-03). That is about realism of the render, not a size promise either.

• Which size the generated garment represents, if it was never told one.
• Real stretch, rise, sleeve length and inseam in centimetres.
• Whether a print is placed the same way on the real piece.
• How the fabric moves when the person walks.

What should an apparel listing carry anyway?

Treat the AI preview as a second layer on top of facts a shopper can check. A size chart, the model's height and the size they wear, garment measurements, and fabric content do the fit work. Keep your own photography as the product's source of truth, because it is the image a try-on feature will be handed as the garment reference.

Sume has no try-on endpoint of its own for consumers and does not publish a fit score. What it has is two catalog Formats, sume-virtual-try-on and sume-virtual-fitting, listed as callable on the Formats by Sume catalog, plus an image API for stills. The fitting Format's own instruction asks for the silhouette, sleeve length, fabric drape and a small turn, which makes a better listing video than a flat lay. It is still a generated preview, so label it as one.

How do you make a fitting preview for a listing?

Call the fitting Format with the garment still as an attachment. The call is one POST to the sume handle; the attachment must be a public HTTPS image, and the body must carry an instruction or input, per Create a Format run. Send an Idempotency-Key derived from your SKU so a retry cannot pay twice, and a spend cap so one run cannot overspend.

Then read the result when the run reaches a terminal status. Put the size chart under the video, not in it.

curl -sS -X POST "https://api.sume.com/v1/formats/sume/sume-virtual-fitting/runs" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: sku-1042-fitting-v1" \
  -d '{
    "instruction": "9:16 virtual fitting preview of the attached shirt. Show silhouette, sleeve length and drape with one small turn. Keep the print exactly as shown.",
    "attachments": [{"type": "input_image", "image_url": "https://cdn.example.com/sku-1042/flat-lay.jpg"}],
    "generation_spend_cap_usd": 20
  }'

What does Sume not do here?

It does not read your size chart, pick a size for a shopper, or certify that the preview matches the real garment. Checking the render against the real piece is a human step, and a short one: compare the print, the colour and the sleeve length with your packshot before publishing. For where the two Formats differ, see the virtual try-on or virtual fitting post.