Product listing image QA

Business value

• Catches the listings where 'product on white' shots were obviously generated, before Amazon/eBay/Etsy listing review catches them and suppresses the SKU.
• Surfaces composite issues — products pasted into impossible scenes, scale mismatches with handheld props — that get flagged in review queues.
• Builds an internal asset-quality bar that scales beyond manual designer review.

Agent job to be done

Be a marketplace listing reviewer. Allow assets that read as actual product photography. Hold assets with composite or generation cues. Reject assets that would obviously fail platform review.

format: product_reviewintended_use: publishdomain: ecommerce product listing photography

When to call VeracityAPI

After product photography/asset upload, before the listing is submitted for marketplace review or syndicated to channels.

What image URL to submit

Public HTTPS image URL for the exact asset that will appear on the listing. Submit the marketplace-resolution version (Amazon main image: 2000x2000 PNG/JPG).

Decision policy

• allow: low synthetic_image_risk, no composite cues, scale and shadow consistency.
• revise: medium risk — request a re-shoot or a different angle from the original photoshoot. Composites can usually be fixed in the next pass.
• human_review: high risk on hero/main images (the photo that decides whether someone clicks the listing).
• reject: visible heavy manipulation on assets that would fail Amazon A+ Content rules or eBay/Etsy authenticity policies.

Automation recipe

• Listing agent receives the asset bundle (hero, secondary angles, lifestyle shots).
• Score the hero image first; if it fails, the rest of the bundle gets paused.
• Allow-tagged listings move to marketplace submission.
• Medium-risk assets enter a designer queue with evidence categories.
• Replacement assets re-enter the gate.

Evidence spans agents should inspect

• 'synthetic_texture' — material textures (fabric weave, leather grain, metallic surfaces) that don't render naturally
• 'composite_seam' — product edges that don't agree with the background lighting or color cast
• 'scale_mismatch' — products that don't make sense at the stated size next to context props
• 'impossible_geometry' — handles, hinges, hardware, or fastener details that violate how the product would actually function

Policy pseudocode

if (result.recommended_action === "allow") continueWorkflow();
if (result.recommended_action === "revise") rewriteWith(result.evidence, result.recommended_fixes);
if (result.recommended_action === "human_review") queueForHumanReview(result);
if (result.recommended_action === "reject") discardOrRebuild();

KPIs to track

• % of listing assets flagged before marketplace submission
• marketplace policy-violation suppression rate (target: drop by 50%+ after deployment)
• designer rework hours per 100 listings
• false-positive appeal rate
• listing time-to-live for clean asset bundles

What can go wrong

• AI-generated product photography is becoming explicitly allowed on some marketplaces (with disclosure) and disallowed on others. The score is one signal; check the platform policy.
• Lifestyle shots are different from product-on-white. Scoring thresholds should differ — lifestyle scenes have legitimate compositing reasons that would trip strict thresholds.
• Pair with a 'has real EXIF camera metadata' check for high-end listings — provenance often resolves ambiguity faster than visual scoring.

Cost and latency notes

Image analysis is a flat $0.02 per image. The endpoint accepts HTTPS image URLs, stores no image bytes, and logs only a URL hash plus hostname. Current v0.1 latency is vision-model-bound, so preflight balance and retry carefully.