Bottle or jar
Scenes: studio background, lifestyle surface, paid social crop
Failure risk: label drift, reflection drift, cap or lid deformation
Benchmark before comparison
AI product photography benchmark template for ecommerce.
Run a small product truth test before claiming an AI product photography tool is good enough for store, marketplace, landing page, or ad use.
Direct answer: A useful AI product photography benchmark tests multiple product categories, multiple scene types, and product truth failures, then records publish, repair, or reject decisions.
Test matrix
Test five product categories across three scene types.
Box or pouch
Scenes: marketplace image, seasonal background, landing page hero
Failure risk: package geometry, typography, fake badges
Transparent or glossy product
Scenes: white background, dark surface, shelf-style scene
Failure risk: transparency, material finish, contact shadow
Apparel or soft goods
Scenes: flat lay, model-free lifestyle, ad variant
Failure risk: fabric texture, scale, color and pattern drift
Accessory or small object
Scenes: macro crop, bundle image, comparison layout
Failure risk: scale, edges, missing parts, invented accessories
Scoring
Score product truth before judging visual quality.
Pretty outputs are not enough. Each image receives two review passes across identity, label, material, geometry, and channel claims before the team decides whether the output can be used commercially.
Identity and variant preserved
Label, logo, and claim text preserved
Material and color truth preserved
Geometry, scale, shadow, and contact points believable
No unsupported badges, discounts, ratings, endorsements, or platform logos
Evidence gallery
Sample evidence records
Use concrete sample records, not empty benchmark claims.
The template includes anonymized ecommerce sample records that show how product truth failures turn into publish, repair, or reject decisions. Treat them as a review-protocol demonstration, not a model ranking.
This is an anonymized review protocol demonstration, not a model ranking.
Sample record: VSK-BM-001
Amber supplement bottle
Prompt input: Premium PDP hero on a light stone surface
Product truth risk: Label drift and cap reflection distortion
Review gate: Check dosage text, cap width, label alignment, and bottle shape
Decision: Réparer
Score: 15/20
Sample record: VSK-BM-002
Matte coffee pouch
Prompt input: Warm kitchen background with seasonal props
Product truth risk: Invented badge and package geometry drift
Review gate: Check roast badge text, pouch corners, seal shape, and product count
Decision: Reject
Score: 10/20
Sample record: VSK-BM-003
Clear skincare serum
Prompt input: Dark reflective surface ad crop
Product truth risk: Transparency, liquid level, and contact shadow drift
Review gate: Check glass transparency, dropper metal, liquid level, and edge halos
Decision: Publish after final channel review
Score: 18/20
Sample record: VSK-BM-004
Sage cotton T-shirt
Prompt input: Model-free lifestyle flat lay with soft daylight
Product truth risk: Fabric color, collar rib, and scale drift
Review gate: Check source color, collar texture, fabric weight, and garment shape
Decision: Réparer
Score: 16/20
Sample record: VSK-BM-005
Measuring spoon set
Prompt input: Marketplace bundle image with clean background
Product truth risk: Included item count and size order drift
Review gate: Count each spoon, confirm nesting order, and reject invented accessories
Decision: Reject
Score: 12/20
Benchmark decision rules
18-20: publish after final channel review
14-17: repair failed details before publishing
0-13: reject or regenerate with narrower edit scope