A product image quality checklist for marketplaces

Marketplace image quality checks should determine whether a photo helps a buyer understand the item and whether it is suitable for the places your product displays it. Resolution is one check. Correct identity, visible condition, and complete framing matter too.

Run checks before editing images so you do not spend money improving the wrong photo.

Start with a technical validation pass

Check that the file is reachable, decodes successfully, has an allowed format, and meets your size and dimension limits. Reject corrupted files and protect processing jobs from unexpectedly large inputs.

Record the actual image dimensions rather than trusting an upload label. Detect duplicate files so the same broken or irrelevant asset does not appear several times in the review queue.

These checks are deterministic and should not need a language model. Reserve visual inference for questions ordinary metadata cannot answer.

Check what the image actually shows

In an illustrative chair listing, the seller uploads a room photo with a chair, a side table, and a lamp. A perfectly sharp image can still leave the buyer unsure which item is for sale.

Ask whether the listed object is present, sufficiently visible, and consistent with the description. If the listing says one chair but the image shows a set, flag the ambiguity. Do not infer that every object in the image is included.

Check text overlays, screenshots, packaging-only images, and photos that show a different variant. A model may help route these cases, but uncertain judgments need a review path.

Use a category-specific checklist

Define a small set of required and recommended views for each category. Do not make sellers upload ten images when three clear views answer the relevant questions.

Separate presentation from condition evidence

A background can be distracting without invalidating the listing. A blurred model label can block identity verification. Route these problems differently.

Use a presentation queue for cropping, background removal, and image ordering. Use an evidence queue for missing damage photos, unreadable identifiers, or contradictions with seller claims. The second queue often needs new seller input rather than an image edit.

See bulk background removal for the checks needed after an edit.

Keep image delivery discoverable

Google's image guidance covers descriptive context, accessible image URLs, and useful alt text. Those are also practical reasons to avoid inaccessible image links and meaningless filenames in a public catalog.

Write alt text that describes the actual image. Do not fill it with search terms or unsupported product claims. Keep important condition information in visible listing content as well; alt text should not be the only place a buyer can find it.

Measure the review burden

Track technical failures, identity mismatches, missing views, and edit rejections separately. An overall pass rate cannot tell you whether sellers need better upload instructions or the processing model needs adjustment.

Sample passed images too. If reviewers inspect only flagged images, you cannot estimate how many problems the system misses. Check results after changing upload guidance, image models, or category rules.

Goodfolk's photo workflow is intended to prepare assets and surface these issues. Start with a category where your team already reviews images. Use the same acceptance criteria for the pilot and compare completed listings, review minutes, and seller follow-up against your current process.