Does an AI product image need a TikTok label? One summary says yes

Yes, according to AuditSocials: its list of label triggers includes photorealistic AI backgrounds and product images (third-party summary, read 2026-10-02). If you make a product shot or a background with an image model, plan to label the post that uses it. That applies whether the image comes from Sume or anywhere else.

What does Sume return for an image?

POST /v1/images takes a prompt, an aspect_ratio such as 9:16 or 4:5, n from 1 to 10, and a free-form metadata object that is stored on the job and not sent to the provider. The result is a media.sume.com URL. The docs page does not describe a label, a watermark or a provenance record added to the file, so assume none is there.

Where does the label decision belong?

On the post, not the image file. Use metadata to carry the decision from generation to publishing, for example the SKU, the channel and a needs_ai_label flag. Your scheduler can then refuse to publish any asset whose flag is true and whose label step is unchecked.

Does the label fix a misleading image?

No. A label says the image is AI-made; it does not make the product look like what you ship. Shop-style rules on accuracy are separate, and the stored posts linked below cover them. Check the shot against the real item before it is labeled and posted.

What about a real product photo with an AI background?

The summary names photorealistic AI backgrounds on their own, so a real product on a generated backdrop is on the list. The same holds when you generate the backdrop and composite the real item over it. The test is whether a viewer would take the scene for a real photo.

Keep the unedited product photo and the generated backdrop in the same folder with the job id. If the listing is questioned, you can show which part was AI-made.

Where can you read Sume's image options?

The full field list is on the Images docs page, and GET /v1/catalog shows which models and ratios your workspace can use today. Check the ratio you need there before a batch, since models differ in the ratios they accept.