← Back to the blog
AI content

AI product photos, what ChatGPT Images 2.6 changes for marketplace sellers

OpenAI has made image generation up to 50% faster. For anyone selling online, the hard part is still keeping the product true to life and following each marketplace's rules.

Photo frames in dark lines on a light background, beside the glass form with a blue core

Good product, weak photo, and the listing falls flat. Anyone who sells on a marketplace knows the scene, which is why every new development in image generation gets attention.

Last week, OpenAI announced ChatGPT Images 2.6. What changes, in practice, for anyone who needs catalogue photos?

What was announced

In the announcement published on 4 November, OpenAI says generation has become 50% faster and presents two models for the API:

  • Flare, for everyday work, which generates images in as little as 6 seconds in the best cases.
  • Sunburst, with higher quality and more consistency between edits.

ChatGPT also now keeps reusable reference images, so there is no need to upload them again. Anyone using the API should check the calendar: some older models, such as gpt-image-1.5, go offline on 1 December.

What the announcement does not say

Speed helps, especially in large catalogues. But the announcement makes no claim of better fidelity to the real product. For a shop, that is exactly the point that prevents returns.

A beautiful photo of a trainer with the wrong sole is a problem. If the colour, label, texture or proportions change, the customer receives something else, and the complaint comes later.

The process matters more than the model

Changing the model changes one stage. The result depends on the whole path:

Five-stage flow from the real product photo to publication, with a check of colour, label and shape before the marketplace rules
  1. Real product photo, with good light and more than one angle.
  2. Generation, with references to keep the item's identity.
  3. Human check of colour, label, shape and details.
  4. Marketplace rules, because background, proportions and text on the image have their own requirements. Always check each channel's seller centre.
  5. Publication and measurement, comparing the listing's performance before and after.

Consistency between edits, the promise of Sunburst, may reduce the rework at stage 3. Only a test with your products shows by how much.

How to test without risking the catalogue

Start with a few products: your best sellers and those with the weakest photos. If the result is good, expand.

That is how our AI catalogue sample works. A few of your products in the studio and in a setting, in your marketplace's format, within 2 working days. On the first order, you see the watermarked sample before paying and approve every piece before delivery.