TikTok Shop main image: no digital renderings, and the real-model shot

TikTok Shop's listing guidance says the main image should not be a digital rendering and that Model Display needs at least one image with a real human model. A fully synthetic product shot sits close to the first exclusion, so the safer use of AI is to edit or extend real product photos, not to replace them.

The two lines that matter

On the seller page the rules for the main image and for Model Display are separate. Both were read on 2026-10-03.

What the page does not say

The page text read here does not mention AI, so it neither allows nor bans generated images. It only supplies the wording above. Treating "digital renderings" as including a fully generated product image is a conservative reading, not a rule the page states, and the platform decides how it enforces it.

That is a reason to avoid depending on the gray area for your main image. A listing that loses its quality tier because the main image was judged a rendering costs more than the shot saved.

A safer split between real and generated

A practical way to use image generation without leaning on the main-image line:

• Use a real photograph of the product as the main image.
• Generate lifestyle and context images from that photograph, passing it as an input_references entry so the product's shape and label carry over.
• Keep the human model image a photograph of a real person who agreed to appear.
• Compare every generated image to the physical product before upload: colors, parts, quantity and text on the pack.

Using a reference image on Sume

The Image API accepts input_references as public HTTPS image URLs. Models whose input_references descriptor is {"min": 0, "max": 0} reject them, so check the catalog first. ChatGPT Image 2.5 (openai/gpt-image-2.5) takes up to 16 references, and Ideogram 4.5 edits the first image with up to four more as references.

On edit calls, prefer aspect_ratio: "auto" to match the reference. The docs note that omitting the field is not the same as sending auto.

{
  "model": "openai/gpt-image-2.5",
  "prompt": "Place this exact product on a sunlit kitchen counter, keep the label unchanged",
  "input_references": [
    { "type": "image_url", "image_url": { "url": "https://example.com/real-product.jpg" } }
  ],
  "aspect_ratio": "auto"
}

Where this leaves a catalog

A mixed gallery of one real main image, one real model image and generated context shots is the lowest-risk reading of the page. It still meets the five-image count, and the two lines about renderings and real models stay untouched.