A product page — and what I refused to let AI make
Evidence12 marketplace slices, 5 teasers and a product video

Evidence12 marketplace slices, 5 teasers and a product video

The detail page for an office chair. Planning, art direction, imagery, graphic design, HTML and marketplace slicing — all mine.
There was no photography budget and no studio. Filling that gap with generative AI is easy, but those pages usually end up showing a product that doesn't match the real thing, or a set of images that don't belong together.
“Bright and friendly, white and mint. Use shots of people seated to build trust.” That line went at the top of the document before anything was generated. Skip it and you get attractive images that refuse to form a page.
The real cut-out product shots stayed; generation was used only for atmosphere. This is an object someone pays for, so its form cannot drift even slightly. Where AI stops isn't a matter of taste — it's a matter of liability.
Generative imagery's worst habit is changing the model and the lighting between shots. I regenerated the three seated shots against the same person, lighting and background until they read as one set that differs only in colour.
16:10 for the hero, 4:3 for seated shots, 1:1 for product shots. Cropping later breaks the composition, so it was decided at generation time.
The layout was finished in PSD and SVG with the image slots left empty, so a disappointing image meant regenerating that one shot rather than touching the layout.
12 marketplace slices, 5 teasers and a product video
AI was the tool; what to make and how far to use it was my call. What I don't have is the conversion data for this page. Next time I'll make receiving the performance numbers part of the handover.
Describe the thing you're currently stuck on. I'll reply by first separating what I can take on from what I can't.