TikTok Shop's nine-image set: white main image, no repeated angles

A TikTok Shop listing takes up to nine images. The main image needs a pure white background and a front view of the physical product with nothing added, and the other eight must each show something different. Sume can clean a background or make a supporting scene from your real photo through POST /v1/images, but the main image should stay a faithful photograph.

Rules here are from TikTok Shop's Product Listing Policy and its page on AI-generated content for listing images, both read 2026-10-02. Check the live pages before a bulk upload, since seller policies change.

What does the policy require for the main image?

The policy page lists the main image rules in plain terms.

• A pure white background and an objective, straightforward representation.
• The front physical view of the product.
• No added logos, text, borders, watermarks or graphics.
• A minimum of 600 x 600 pixels.
• In color; black-and-white is not acceptable.

What about the other eight images?

Additional images, up to nine in total, should show front, back, sides and other details, including accessories. They may show key features, usage scenarios, variations, styled scenes, close-ups and size comparisons. Two rules matter for batch work: the same 600 x 600 minimum and no overlays apply, and multiple images showing the same angle are not allowed.

That last rule is where a naive bulk run fails. Asking a model for nine lifestyle variations gives you nine versions of the same angle. Plan the angles first, then generate one image per slot.

Where does AI help, and where does it break the rule?

The AI page says AI-generated content is allowed when it accurately represents the real product, and prohibits changing size, color, material or shape, showing features or performance the product lacks, and misrepresenting bundles or quantities. So Sume is a fit for slot 1 background cleanup and slot 7 styled scenes, where the product is a fixed reference, and a poor fit for slots 2, 3, 5, 8 and 9, which are claims about what the buyer receives.

Do not use a generated view to fake a back or side you never photographed. If the model invents a button or a seam, you have shown a feature the product does not have.

How do I call Sume for slots 1 and 7?

Send your photo as an input_references image with an edit prompt. The images docs say to prefer aspect_ratio: "auto" on edit calls so the output matches the reference, and that reference URLs must be public HTTPS. Read supported_parameters from GET /v1/images/models before pinning a model or tier.

Always open the result next to the original and compare the logo, the proportions and the count of items before you upload it.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/images",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}",
             "Idempotency-Key": "nine-set-sku-1042-slot-7"},
    json={
        "model": "openai/gpt-image-2",
        "prompt": "Place this exact product on a kitchen counter in morning light. "
                  "Do not change its shape, color, logo or size.",
        "input_references": [{"type": "image_url",
                              "image_url": {"url": "https://example.com/sku-1042-front.jpg"}}],
        "aspect_ratio": "auto",
    },
    timeout=60,
)
r.raise_for_status()
print(r.json())

Should I use n to get all nine in one call?

The images docs let n ask for 1 to 10 images per call, but note that per-model ceilings are lower, and one prompt with n gives variations of one idea, which is the opposite of what this policy wants. Nine different angles are nine different prompts.

Billing is all-or-nothing per generation: a completed generation is billed in full, and a failed or cancelled one is not billed. So a failed slot-7 call costs nothing, and you can re-run just that slot.

Resolution matters too. The listing minimum is 600 x 600 pixels, which the 1K tier clears, but check the width and height of what comes back instead of assuming, especially after an aspect_ratio: "auto" edit of a small source photo.

How do I keep the set consistent across SKUs?

Keep one prompt per slot and change only the product URL. Review the output per slot, not per SKU, so a bad slot-7 prompt is fixed once. The related post on a white-background edit covers slot 1 in more depth, and the four-shot Walmart plan shows the same one-image-per-slot idea for another marketplace.