eBay variation pictures: one picture set per color, not per SKU

How do eBay variation photos work?

On eBay you do not attach a photo to each variation row. You pick one variation-specific name, such as Color, as the classifier, and then give each value of it its own picture set: a set for Blue, a set for Black, a set for Pink. Every blue variation, whatever the size, shows the blue set. That comes from the Picture rules in eBay's Multiple-variation listings guide, read 2026-10-03.

For an image pipeline this changes the unit of work. A listing with 3 colors and 5 sizes is 15 variations but only 3 picture sets, so you generate per color, not per SKU row. With 5 shots in each color set that is 3 x 5 = 15 images, not 15 x 5 = 75.

What limits does eBay put on each set?

eBay's guide says each variation picture set can hold the same maximum number of pictures allowed at the item level for that marketplace and category, and its example cites up to 12 eBay Picture Services pictures. Check the cap for your own marketplace and category before you size a batch.

Hosting matters too. The guide says all pictures in the item-level set and the variation sets must use the same type of picture hosting, either eBay Picture Services or self-hosted pictures, and you cannot mix both in one listing. Sume returns Sume-hosted URLs, so from eBay's side they are self-hosted pictures. If the rest of the listing uses eBay Picture Services, upload your Sume results there too, rather than mixing.

How do I generate one set per color with Sume?

Start from one clean photo of the real product, since eBay's rules are about showing the item itself and your recolors must match the stock you sell. Sume's Image API takes up to 5 images on Ideogram 4.5 in total (the first is edited, up to 4 more are references), and on an edit aspect_ratio: "auto" keeps the source shape; the docs note that omitting the field is not the same as auto.

The request below recolors the base photo for one value of the classifier. Run it once per color, then add the angle shots for that color the same way.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ideogram/ideogram-v4.5",
    "prompt": "Same sneaker recolored to blue, white studio sweep, side view",
    "aspect_ratio": "auto",
    "quality": "medium",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/sneaker-base.jpg"}}
    ]
  }'

What should I check before publishing?

Compare each generated color against a real unit. A generated pink that is a shade off is a returns problem, not an image problem. Then count: sets per classifier value, pictures per set against the cap, and one hosting type across the listing.

Image generation on Sume is billed only when it completes; failed or cancelled generations are not billed, and slow requests can return 202 with a job to poll instead of the image body. Treat each color group as its own retryable unit so one slow color does not block the others.