Rakuten Ichiba image rule: text on 20% or less of the photo

Rakuten Ichiba's Product Image Registration Guideline says text elements inside a product image must occupy 20% or less of it, and it covers the first product image and the SKU images. The background must be single white or a photographic background, frames and borders are not allowed, animated GIFs are not allowed, and Rakuten recommends square images of 700x700 pixels or larger in JPEG or PNG. That comes from Rakuten's RMS Service Square column, read on 2026-10-03; the page was published in March 2018 and shows an update of December 26, 2025.

What does the guideline cover?

It targets the first product image and the SKU image in the product image section of the store tool, The page names no other image slots, so read the guideline as limited to those two.

How do I make a text-light image on Sume?

The simplest way to stay inside 20% is to generate the first image with no text at all and add the product name or a short catchphrase afterward in your own layout tool, sized to a small share of the frame. Ask the model for a clean product on white or on a natural photographic setting, and say explicitly: no border, no frame, no overlay text, no logo banner.

On Sume, openai/gpt-image-2.5 takes a custom image_size with both edges as multiples of 16, between 655,360 and 8,294,400 pixels. 700x700 itself does not qualify (700 is not a multiple of 16), and the nearest multiple, 704x704, is only 495,616 pixels, below the minimum. 1024x1024 clears both, and it is larger than the 700 pixels Rakuten recommends. Ask for output_format: "jpeg" so the file is a type Rakuten lists; the Image API accepts png, jpeg, webp or svg.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Studio photo of this product centred on a plain white background. No text, no border, no frame, no logo banner.",
    "image_size": "1024x1024",
    "output_format": "jpeg",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/product.jpg"}}
    ]
  }'

How do I check the 20%?

Measure after you add text, not before. Open the final file, mark the area each text block covers, and add them up against the whole frame. If the product name and a badge together cover a fifth of the picture or more, cut them, or move that information to the second image. The Rakuten page states the limit; it does not describe a scoring method in the part I read, so use a conservative margin rather than aiming for exactly 20%.

Frames are the other quiet failure. Image models often add a thin border or vignette on request for a catalogue look, so name "no border" in the prompt and look at the edges of the result. Animation is not an issue for stills, though a GIF or video-derived frame from another workflow is: export a still instead.

How does this compare with other marketplaces?

The marketplaces disagree about what belongs on the first image. Coupang's main image wants a white background and a product filling most of the frame, and Walmart bans seller logos where Coupang wants yours. Keep one prompt per marketplace and never reuse an export between them: a Rakuten first image that carries a badge is a rejection risk the moment it is reused somewhere with a stricter rule.

What should the SKU images follow?

The guideline names SKU images alongside the first product image, so a listing with colour or size variants has more than one picture under the 20% limit. Generate each variant from its own reference photo, with the same prompt wording, so the framing and background match across the set. Keep the prompt and the output file name per SKU in a spreadsheet, so a rejected image can be regenerated without touching the others. If Rakuten asks you to resubmit one, change only that file.