How to make AI photos for beauty and skincare products
Use AI for packs, textures and mood, and keep it away from results. Models render cream swirls, serum drops and frosted glass well, but regulators treat a picture of skin as a claim about what the product does, however the picture was made. Never use an AI face as an "after" shot, and match makeup shades to the real product with an eyedropper.
What AI does well and badly for beauty
| Subject | How AI does | Watch for |
|---|---|---|
| Texture swatches: cream, gel, serum, balm, powder | Well | A thickness or sheen that is not your formula |
| Packaging: frosted glass, soft-touch tubes, metal caps | Well, from good references | Text on curved jars and tubes; caps reflecting an invented room |
| Ingredient still lifes | Well | Ingredients the product does not contain |
| Skin close-ups | Poor | Pores and texture smoothed away; a result the product cannot give |
| Makeup on skin | Poor for accuracy | Shade drift; coverage the product cannot give |
| Hair | Mixed | Volume and shine beyond what the product does |
| Hands applying product | Poor | Fingers, nails, and the amount of product on them |
The line runs between objects and skin. A jar or a swatch shows what the product is. A face or a patch of skin can show what the product does, which turns the picture into a claim and brings in the rules further down this page.
How to brief a texture swatch
A swatch shot works only if the formula looks like your formula. Photograph a real swatch first: a smear on glass, a dollop on a spatula, a drop on skin. Then describe what the camera should see.
| Formula | Words that describe it |
|---|---|
| Rich cream | "thick cream holding soft peaks, satin sheen, ridges left by the spatula" |
| Gel | "clear gel with a wet shine, trapped micro-bubbles, edges that slump slowly" |
| Serum | "a single drop hanging from a glass dropper, clear with a faint gold tint (#E9D9A6)" |
| Balm | "dense balm with a matte top and one clean scoop mark" |
| Pressed powder | "pressed powder with a fine crumble at the break, matte, no shine" |
| Lipstick | "bullet with a crisp angled tip and a satin finish" |
Add the swatch color as a hex code sampled from your own photo, and give one light direction. Raking side light shows texture; soft frontal light hides it.
Packaging finishes that models get wrong
- Frosted glass: soft, wide highlights and no sharp reflections. Say "frosted" and "diffuse highlight," or the model gives you clear glass.
- Metal caps and pumps: chrome and gold mirror their surroundings. Say what they reflect, or keep the set plain.
- Soft-touch plastic: matte, with a faint sheen only along the edges.
- Clear serum bottles: the liquid bends what is behind it, and a label on the back shows through reversed.
- Tubes: crimped ends and text that curves with the tube. Treat tube text like can text; see keeping labels readable in AI photos of packaged goods.
Skin, faces and casting
Image models often smooth skin, which looks fake and, in a skincare ad, looks like a result. Ask for visible pores, fine lines where the person’s age would have them, and uneven tone in natural light.
Cast deliberately. In a 2023 note on generative AI, CAP, which writes the UK advertising code, warned that some tools tend to show people in higher-paying jobs as men or as lighter-skinned, and to portray "idealised body standards," and asked advertisers to check that AI images do not carry harmful stereotypes. Render each skin tone from its own reference, and check that your shade looks right on each one. See how to cast AI models responsibly.
Some countries require labels. In France, a June 2023 law on influencers requires the words "Images virtuelles" on influencer content that uses AI to depict a face or a body, and "Images retouchées" when a face or body shape has been edited.
The rules on showing results
UK guidance is specific about images. CAP’s advice on production techniques says visual claims "should not misleadingly exaggerate the effect the product is capable of achieving." It names the usual problems: lash inserts that do more than the mascara could, retouching the very area the product treats (lines around the eyes in an eye cream ad), and filters. In 2021 the ASA ruled against two tanning ads because the influencers’ filters changed the skin tone the product was meant to change. In 2023, CAP added that AI images used to show efficacy can mislead "in the same way as images that are photoshopped or subjected to social media filters might."
Before-and-after pictures get the strictest reading. CAP treats them like testimonials: you need signed and dated proof that the photos are genuine and unaltered, and the "after" should show what customers can generally get. The ASA has called such shots "objective visual claims" of a product’s efficacy. In September 2026 it ruled against ads for an AI appearance app because nothing showed the before-and-after footage was genuine, used the same people, or was typical. An AI-made "after" face shows no one’s real result, so it cannot pass that test.
In the EU, the cosmetics claims regulation (655/2013) covers claims made through "pictures and figurative or other signs," and says a product’s performance must not be presented beyond the supporting evidence. In the US, FDA decides whether a product is a cosmetic or a drug by its intended use, which claims in labeling and advertising establish; promises to reduce cellulite or regenerate cells make a product a drug. A picture that shows cellulite vanishing makes the same promise, so treat it with the same care.
| Image | Risk | Safer version |
|---|---|---|
| AI "after" face in a skincare ad | Shows a result no real customer got | Real customer photos with signed, dated proof, or no before-and-after |
| AI model with flawless skin beside a serum | Reads as the serum’s result when skin is what it treats | Visible skin texture and product-focused framing |
| Mascara on AI lashes | Length and volume the mascara cannot give | Real application photos with no inserts beyond replacing missing lashes |
| Tan or bronzer shade on AI skin | A color change the product cannot make | Real application photos in neutral light |
| Texture swatch or pack on a set | Low, when the formula and pack match | Keep them matched to the real product |
Getting makeup shades right
- Photograph real swatches of every shade in daylight-balanced light, with a gray card in the frame.
- Sample each shade with an eyedropper and record the hex code next to the shade name.
- Put the hex code and the shade name in the prompt, and send the swatch photo as a reference.
- Check the render with the same eyedropper. If the shade drifted, correct it in post or remake the image.
- Show each shade on a skin tone it is made for, rendered on its own, instead of recoloring one render into the others.
A shopper who picks a shade from a picture expects that color when the box arrives. For color checks and screen versus print color, see how to keep brand colors accurate in AI images.
Checks for beauty images
- Pack text, cap and finish match the real product
- Formula color, thickness and sheen match a real swatch
- Shade hex codes match your recorded values
- Skin keeps visible texture, and the area the product treats was not smoothed or recolored
- No before-and-after was made or improved with AI
- Lashes, hair and nails show only what the product itself can do
- No ingredient appears that the product does not contain
- Each skin tone was rendered from its own reference and checked
- Influencer content in France that uses AI faces or bodies carries "Images virtuelles"
- Claims in words and pictures match your evidence
Questions people also ask
- Can I use an AI model to show how my foundation looks on skin?
- As a mood image, with care. As proof of coverage or finish, no: the picture is then a claim about what the product does, and a generated face never wore it. Use real application photos for coverage and finish, and keep AI for packs, textures and scenes.
- Can I retouch AI images of skin?
- General cleanup is normal, and CAP says minimal airbrushing is likely to be acceptable when it does not imply the claimed effect. Do not retouch the area your product treats: lines for an eye cream, pores for a pore serum, lashes for a mascara.
- Do I have to label AI images in beauty ads?
- It depends on where and how you advertise. France requires "Images virtuelles" on influencer content that uses AI to show a face or body, and ad platforms run their own AI labels. See whether you need to label AI images in ads.
Where Overs fits
Overs can cast packaged products such as skincare: you approve a character sheet of one person before any other photo is made, and the same face carries through the set. On Pro and Team plans, a reviewer checks each photo against its references before you see it. Whether an image makes a claim you can support is still your call.
Free for 40 photos a month. The AI that makes the photos is billed separately, on your own key, with no markup from Overs.
Sources
- ASA and CAP: Beauty and cosmetics, the use of production techniques (Aug 18, 2025)
- ASA and CAP: Before and after photos (Jun 5, 2025)
- ASA: Ruling on Oneade (May 4, 2022)
- ASA: Ruling on Glow Up LLC t/a Maxxing (Sept 2, 2026)
- ASA and CAP: Generative AI and advertising, decoding AI regulation (Aug 31, 2023)
- EUR-Lex: Commission Regulation (EU) No 655/2013, common criteria for cosmetic product claims
- Légifrance: Loi n° 2023-451 du 9 juin 2023 on commercial influence (article 5)
- FDA: Is it a cosmetic, a drug, or both? (or is it soap?)