# AI Product Photography Product Truth Scorecard

Version: 2026-07-06
Source: Visual Skill Kit workflow framework

## What This Scorecard Is

This is a review protocol for AI-assisted ecommerce product images. It is not a model benchmark and does not claim that one generator is better than another.

Use it when an AI tool creates or edits a product image from a real product reference. The goal is to decide whether the image can be published, needs local repair, or should be rejected.

## The Rule

Product facts are fixed. The scene is editable.

Fixed product facts:

- Product identity
- Variant, pack size, and included accessories
- Shape, geometry, and proportions
- Label text and logo placement
- Material, color, texture, transparency, and finish
- Visible claims, badges, certifications, prices, and offer details

Editable layer:

- Background
- Surface
- Lighting
- Crop
- Composition
- Negative space
- Campaign mood
- Channel layout

## Review Method

Compare the AI output against the approved source image at full size.

Score each item:

- 2 = Pass
- 1 = Needs repair
- 0 = Reject

If any product fact scores 0, do not publish the image until repaired or regenerated from a safer workflow.

## Product Truth Tests

| Test     | Question                                                                        | Score     |
| -------- | ------------------------------------------------------------------------------- | --------- |
| Identity | Would the buyer recognize the exact same product and variant?                   | 0 / 1 / 2 |
| Shape    | Are proportions, silhouette, edges, and visible parts unchanged?                | 0 / 1 / 2 |
| Label    | Are label words, typography, and placement still faithful?                      | 0 / 1 / 2 |
| Logo     | Is logo geometry, color, and placement unchanged?                               | 0 / 1 / 2 |
| Material | Does the finish still match the real product?                                   | 0 / 1 / 2 |
| Color    | Are color, variant, transparency, and texture truthful?                         | 0 / 1 / 2 |
| Scale    | Does the product scale make sense against props or hands?                       | 0 / 1 / 2 |
| Shadow   | Do contact shadows and reflections support the product instead of deforming it? | 0 / 1 / 2 |
| Claims   | Were no fake claims, badges, reviews, discounts, or certifications added?       | 0 / 1 / 2 |
| Channel  | Would the image pass store, marketplace, ad, or landing-page review?            | 0 / 1 / 2 |

## Decision Rules

| Total | Decision                                      |
| ----: | --------------------------------------------- |
| 18-20 | Publish after final channel review            |
| 14-17 | Repair the failed details before publishing   |
|  0-13 | Reject or regenerate with narrower edit scope |

Automatic reject:

- Product variant changed
- Label or logo rewritten
- Fake certification, rating, price, claim, endorsement, or marketplace badge added
- Product size, included accessories, or material misrepresented
- The image implies a buyer will receive something different

## Common Failure Modes This Catches

1. Label drift: text becomes plausible but wrong.
2. Logo drift: logo shape changes or moves.
3. Variant drift: color, size, scent, flavor, or pack count changes.
4. Material drift: matte becomes glossy, fabric becomes plastic, metal becomes glass.
5. Geometry drift: edges warp, product becomes taller, thinner, or asymmetrical.
6. Scale drift: props make the product look larger or smaller than reality.
7. Shadow drift: contact shadow makes the product float or melt into the surface.
8. Claim drift: AI adds badges, stars, discounts, awards, certifications, or unsupported text.

## Safe Prompt Starter

Use the uploaded product photo as the fixed source of truth. Create an ecommerce product image for [channel] with [background or scene direction]. Preserve product shape, proportions, label text, logo placement, color, material, texture, edges, contact shadow, and visible claims. Change only the background, lighting, surface, crop, and composition. Do not invent badges, ratings, certifications, discounts, accessories, or product features.

## Citation Note

When citing this scorecard, describe it as a workflow review protocol for AI product photography, not as an empirical model benchmark.