← ClaudeAtlas

product-image-qa-reviewlisted

Review generated product images before they ship, catching the artefacts and accuracy failures that pass casual inspection. Use whenever the user is about to publish product images, has a batch to check, or asks how to quality-check AI-generated product photography.
apimageorg/apimage-skills · ★ 0 · Code & Development · score 72
Install: claude install-skill apimageorg/apimage-skills
# Product Image QA Generated product images fail in ways that are invisible at thumbnail size and obvious to a customer holding the item. A warped logo, an invented rear panel, a shifted colour — all pass a glance and none pass a return. This is the review pass. It takes a few minutes per image and it's the difference between using generated imagery commercially and getting away with it until you don't. ## The two categories of failure **Artefacts** — the image is wrong. Warping, garbling, halos, extra objects. Aesthetic, embarrassing, and usually easy to spot once you know where to look. **Accuracy failures** — the image is *plausible but not your product*. Shifted colour, invented detail, fabricated text, a configuration you don't sell. Harder to spot and much more consequential: these drive returns, complaints and misrepresentation exposure. **Accuracy is the one to prioritise.** A slightly soft edge costs nothing. A navy shirt rendered as charcoal costs a return, a refund and a review. ## The review pass ``` ACCURACY — against the real product, not the reference photo [ ] Colour matches the physical product under neutral light [ ] Shape and proportions correct [ ] Label copy: correct, or invented? [ ] Logo: exact shape, or re-drawn? [ ] Materials and finish read correctly [ ] Any face the references didn't cover — fabricated? [ ] No features added that don't exist [ ] No configuration or variant you don't sell [ ] Nothing in frame that doesn't ship with it ARTEFA