Skincare is the hardest thing to shoot with AI, and it is the category most likely to be sold on a phone screen with a generated image. A serum bottle is frosted or clear glass, it has a dropper and a metallic or matte cap, it throws reflections, and it carries a small label with a brand name and claims that have to come out right. Every one of those is a place an image model can quietly drift.
So I tested it. I ran one skincare brief, a frosted glass dropper bottle labeled "AURELIA" on wet concrete with a sprig of eucalyptus, through four of the strongest image models with the same prompt: Nano Banana 2, GPT Image 2, Seedream 4.5, and FLUX.2 Pro. The results split cleanly, and the split is useful: two models win on label fidelity, one wins on the premium hero look, and the cheapest one garbled the brand name. This is the skincare companion to our best AI image model for product photography roundup.
Almost every "AI for skincare photos" guide points you at a tool, Photoroom, Claid, Omi, Nightjar, and most of them are right that fidelity is the whole game for skincare. But those tools are wrappers around the same handful of image models, and several hide which model they run so you inherit one product's house style. This post is one level down: the actual models, tested on the actual hard case, so you can pick the right one instead of guessing through a wrapper.
Quick answer: which model for which skincare shot
- The full label and claims have to be legible (PDP hero, anything with regulated copy): GPT Image 2, with Nano Banana 2 close behind. They held the small print.
- A premium lifestyle hero where only the brand name shows: Seedream 4.5. The most cinematic frosted-glass look, and the best value of the photoreal options.
- Cheap volume and you will overlay the label yourself: FLUX.2 Pro. Cheapest and editorial, but it malformed the brand letters, so do not trust its text.
If you only remember one thing: glass and the dropper were not the problem for any model. The label was. Test your own bottle for label fidelity before you trust a model with a line.
The test, model by model
One brief, four models, same prompt. Here is what each produced and where it held or broke.
Nano Banana 2 gave the most balanced result. The frosted glass reads right, the dropper and matte cap are faithful, the contact shadow and the reflection in the wet surface are believable, and crucially the label kept its brand name and the smaller line beneath it. Nothing flashy, nothing broken. For most skincare work this is where I would start.
GPT Image 2 is the one to reach for when the words on the bottle matter. It rendered the most label detail of any model, adding a legible product line and fine print that the others either simplified or dropped, on a brighter studio-lit backdrop. The tradeoff is the one you would expect: it is the most expensive per image and the slowest to return. For a regulated cosmetic label or an ingredient callout, that tradeoff is usually worth it.
Seedream 4.5 made the most beautiful image. The frosted-glass texture, the eucalyptus reflected in the water, the depth of field, it looks like a paid editorial shot, and at roughly half the cost of Nano Banana 2. The catch for skincare specifically: it kept the brand name clean but dropped the smaller line of label text. For a lifestyle hero where only "AURELIA" needs to read, that is fine and the look is the best here. For a PDP image that has to show the product line or volume, it is not.
FLUX.2 Pro was the cheapest and gave the most editorial, minimalist composition, warm gradient backdrop, the amber serum visible through the glass, scattered water droplets. It is genuinely pretty. But the label is where it falls down: the brand letterforms are subtly broken, the kind of malformed type that looks fine in a thumbnail and wrong the moment a customer zooms in. This matches what FLUX does elsewhere, strong main image, unreliable incidental text. Use it when you plan to overlay the label yourself.
The comparison
| Model | Best for | Label fidelity | Hero look | Rough cost/image | Watch out for |
|---|---|---|---|---|---|
| Nano Banana 2 | Balanced default for skincare | Strong, kept brand + sub-text | Clean, photoreal | ~9.3 credits | Nothing major; verify at 4K |
| GPT Image 2 | Legible claims and label copy | Best, most detail and fine print | Bright studio | ~26.4 credits | Priciest, slowest |
| Seedream 4.5 | Premium lifestyle hero, value | Brand clean, dropped sub-text | Best here | ~4.8 credits | Loses small label text |
| FLUX.2 Pro | Cheap volume, label overlaid | Weak, malformed brand letters | Editorial | ~3.6 credits | Garbles incidental text |
Credit costs are first-hand from this test on Masonry; per-image rates move, so check current pricing before you budget.
Why skincare is harder than most product photography
A matte cardboard box is forgiving. A serum bottle is not, for three reasons that all showed up in this test.
Glass and reflections. Frosted and clear glass force the model to invent what is behind and around the bottle: the fill line, the way light scatters through the frost, the reflection on a wet surface. All four models actually handled this well, which is the good news. Glass is no longer the hard part.
The label is the hard part. Cosmetic labels carry a brand name, a product line, a volume, sometimes ingredient or claims text. That small printed type is exactly what image models smear, and it is exactly what a skincare customer reads. This test made the divide clear: GPT Image 2 and Nano Banana 2 held it, Seedream simplified it, FLUX broke it. If your label carries regulated copy, that is not a detail, it is the whole shot.
Trust and returns. Shoppers have gotten wary of AI imagery. In Deloitte's 2025 Connected Consumer survey of 3,524 U.S. consumers, 59% said they cannot reliably tell AI-generated content from real, and most want it disclosed. For skincare that cuts two ways: a shot that obviously reads as AI erodes trust right as someone decides to buy, and a shot that flatters the product into something it is not sets up a return. A faithful bottle, with the real label, protects the sale and the margin.
How to shoot your skincare line without a studio
The workflow is the same one the product photography roundup lays out, applied to the hard case. Start from a clean reference of your actual bottle. Run it through two models, not one, because the model that nails your frosted serum may butcher your foil-stamped cream. Judge every output for the label first and the look second. Proof the final at full size, and for any regulated claim, overlay the real copy yourself rather than trusting a from-scratch render.
With the Masonry CLI you can fire the same skincare prompt at every model from one command and compare the outputs side by side, which is exactly how the four images above were made:
masonry image "frosted glass serum bottle, AURELIA label, eucalyptus, wet concrete, soft studio light" --model gpt-image-2 masonry image "frosted glass serum bottle, AURELIA label, eucalyptus, wet concrete, soft studio light" --model seedream-4-5
The bottom line
Skincare is the category where the model choice actually matters, because the label is the product and the label is what models get wrong. For claims and legible copy, GPT Image 2 or Nano Banana 2. For a premium hero where only the brand shows, Seedream 4.5 at a third of the price. FLUX.2 Pro for cheap, pretty, label-free volume. The only real mistake is trusting one model's demo over a test on your own bottle. Run yours across two from one place with the Masonry CLI, or see how the same logic plays out across all product types in our best AI image model for product photography roundup.
