Score AI product photos before publishing. Check whether the output still represents the exact product a buyer will receive: identity, label, logo, material, color, scale, claims, and channel risk.
Use this after generation and before upload. It is not a style score; it is a buyer-truth score.
Compare the AI output against the approved source image at full size. Score truth before style, because a polished product image can still be commercially unsafe.
Would the buyer recognize the exact same product and variant?
Variant, pack size, included accessories, or core product identity changed.
0 / 1 / 2Are silhouette, proportions, edges, and visible parts unchanged?
The product looks taller, wider, thinner, warped, melted, cropped, or rebuilt.
0 / 1 / 2Are label words, typography, and placement still faithful?
The model rewrote label text, changed ingredient or warning copy, or invented text.
0 / 1 / 2Is logo geometry, color, and placement unchanged?
The logo moved, stretched, disappeared, or became a lookalike mark.
0 / 1 / 2Does the finish still match the real product?
Plastic becomes glass, matte becomes glossy, fabric texture changes, or transparency drifts.
0 / 1 / 2Are color, variant, transparency, and texture truthful?
The image implies a colorway, variant, or material the buyer will not receive.
0 / 1 / 2Does product scale make sense against props, hands, or packaging?
The product looks larger, smaller, or differently proportioned than the real item.
0 / 1 / 2Do contact shadows and reflections support the product instead of deforming it?
The product floats, bends at the edge, gets merged into the surface, or casts impossible shadows.
0 / 1 / 2Were no fake claims, ratings, badges, prices, certifications, or endorsements added?
The image adds review stars, certifications, discounts, marketplace badges, or unsupported claims.
0 / 1 / 2Would the image pass store, marketplace, ad, or landing-page review?
The image could mislead a buyer, violate a listing rule, or require disclosure before use.
0 / 1 / 2Keep source image, prompt, output, and reviewer notes in the asset record.
Use local repair or regenerate with a narrower edit scope. Do not publish until failed tests pass.
The output changed too many buyer-facing facts. Return to the approved source image and simplify the job.
This block is written to be cited. It gives the method first, without requiring a tool ranking or benchmark claim.
A product truth scorecard is a review checklist for AI product images. It compares the generated output against an approved source image and scores buyer-facing facts such as identity, label text, logo, material, color, scale, claims, and channel risk.
Use it before publishing AI-generated product photos to a store, marketplace, ad, landing page, catalog, or social campaign. It is most useful when AI changed the scene, background, lighting, crop, or layout around a real product.
No. It is a workflow review protocol, not an empirical benchmark. It does not claim that one AI image model is better than another; it helps teams decide whether a specific output is safe enough to publish.
Fail the image if product identity, variant, label text, logo, material, color, scale, included accessories, or visible claims changed, or if the image adds fake ratings, certifications, prices, endorsements, discounts, or marketplace badges.