AI Before and After Videos for Skincare: ASA Compliant Patterns

The ASA upheld more skincare complaints in 2025 than in any year of its existence, and the single most-banned asset format was the before-and-after shot. AI before and after videos for skincare brands solve a real problem (variants are cheap, models are not), but they also industrialise the exact format the regulator is hunting. Most brands using AI here are walking into the trap with their eyes closed.

Generating a compliant before-and-after is harder than generating a generic product shot. The regulator's expectations are specific. The script does most of the work. The footage matters less than people think.

This is what the ASA actually polices, and how DTC skincare brands are building AI before/after pipelines that don't get banned in week three.

Why before/after sits at the top of the ASA's hit list

The CAP code's section 12 governs medicines, medical devices, and health-related products. The Committee of Advertising Practice has issued repeated guidance on cosmetic advertising over the last five years; the consistent thread is that before/after imagery makes implicit efficacy claims, and implicit efficacy claims need the same level of substantiation as explicit ones.

A photo of clear skin next to a photo of broken-out skin, with your serum bottle in between, is making a claim. The implied claim is "this product produced this transformation." That claim has to be supported by evidence the regulator finds adequate. "Adequate" in skincare means clinical study data on the actual product (not the lead ingredient in isolation), with controls, reasonable sample size, and methodology you'd be willing to publish.

Most DTC skincare brands don't have that data on most of their SKUs. So the before/after format becomes a regulatory bet. The ones the ASA goes after share a few features: dramatic transformations, short timelines, and any acne, eczema, rosacea, or dermatitis context. The ones that survive are the ones that frame the result modestly and stay away from the medical-condition adjacency.

ASA's CAP code section 12 is the canonical source. The bit DTC marketers underestimate is that "puffery" defences that work in fashion don't work in skincare. The regulator treats skincare claims as health-adjacent and applies a higher bar.

What ASA bans (and what it doesn't)

A non-exhaustive but useful list of what gets pulled down.

Banned almost without exception:

• "Cleared my acne in 7 days" (timeline + medical-condition framing + transformation claim)
• "Cured my rosacea" (treatment claim on a cosmetic)
• Filter-edited or AI-generated dramatic transformations passed off as real results (the AI angle does not mitigate, it aggravates)
• Before/after with dermatologist-style commentary if the speaker isn't a registered dermatologist
• Any "results after one use" claim on a product that doesn't have clinical data showing one-use results
• Comparing your serum to retinol prescriptions or acne medications

What survives:

• Modest, realistic improvements over realistic timelines (8-12 weeks for most actives)
• Disclosed AI generation when the asset is synthetic
• Before/after where the "after" is described as the customer's experience rather than a guaranteed outcome
• Substantiated claims about hydration, barrier support, evenness of tone, and texture, where you have data on YOUR product
• Realistic shooting conditions (the same lighting, makeup, and camera angle on both sides)

The pattern: the ASA polices the gap between what the visual asset implies and what the brand can prove. AI compresses the cost of producing the asset but does nothing to close the proof gap. That's the problem to solve.

Prompt patterns that produce compliant before/after videos

Three patterns we've seen work for AI-generated skincare before/after assets that don't get banned. These assume you have substantiation for the claim level you're making; if you don't, no prompt will save you.

Prompt 1, hydration improvement, 4-week framing

Female 30-something, bare face, soft natural daylight, kitchen window. Show two camera angles intercut: "before" frame is morning routine, slight texture and dullness visible (realistic, not exaggerated). "After" frame, same lighting, same angle, four weeks later, captioned "4 weeks of using [product]". Improvement is subtle: more even tone, slight luminosity, less visible dryness. Voiceover: "I'm not promising miracles, but my skin actually feels different." Avoids any claim about treating conditions. No prescription comparisons. No "after" frame with makeup if "before" is bare face. Disclose AI in caption.

Prompt 2, texture refinement, 8-week framing

40-something speaking to camera in bathroom mirror, two recordings side by side. Left frame is week 1, right frame is week 8. Lighting and posture matched. Discusses texture and how the product fits into their morning routine. Mentions they use it alongside SPF and a hydrating serum. Result described as "smoother, more even" rather than "transformed." No mention of acne, breakouts, or skin conditions. Tone honest, slightly self-deprecating. Disclose AI generation.

Prompt 3, even tone, 12-week framing for hyperpigmentation-adjacent products

Mid-30s woman, neutral background, soft lighting, no makeup. Two takes: week 1 and week 12. Acknowledge the timeline upfront in voiceover ("I've been using this for three months"). Result is modest evenness improvement. AVOID: any mention of melasma, post-inflammatory hyperpigmentation, dark spots framed as a medical concern. STAY: in the cosmetic envelope of "evenness," "tone," "luminosity." Disclose AI.

The common thread: long timelines, modest results, no medical-condition vocabulary, matched shooting conditions, AI disclosure. None of these are harder to brief than a non-compliant version. They just require knowing what to avoid before you write the brief.

Why AI changes the cost of compliance, not the substance

A traditional skincare before/after shoot, done compliantly, costs £4,000 to £12,000 across talent, production, and the multi-week timeline you actually need to capture before/after honestly. Most DTC skincare brands fake the timeline (shoot before/after in one session with different lighting and makeup) which is exactly what the ASA is increasingly catching.

AI before/after generation costs roughly £8 to £30 per finished asset at scale. The economics flip: you can afford to generate eight variants per claim level, A/B test them on Meta, and only invest in shooting the winner. The compliance work moves upstream into the brief; the production work disappears.

The trap is that AI also makes it cheap to generate non-compliant variants. Brands that scale this category badly end up with thirty banned ads in a quarter. Brands that scale it well treat AI as a bigger funnel that needs a tighter compliance check at the top, not the bottom.

Cost-per-AI-video data by model in 2026 breaks down which models hit which price points, if the cost piece is the part you're modelling.

How a vertical-aware platform handles ASA pre-flight

General AI video tools have no idea what the ASA is. Tonic's skincare vertical has a pre-flight that flags the specific phrases and visual patterns the ASA goes after: medical-condition vocabulary in scripts, unrealistic timelines in captions, dramatic transformation cues in shot descriptions. It also enforces a "matched conditions" check on before/after briefs (lighting, angle, makeup state) so the asset doesn't lie by setup.

This is not a substitute for substantiation. If you don't have data on your product, you cannot make claims at any compliance level. What the pre-flight does is stop you from accidentally over-claiming on data you do have. Most banned skincare ads don't fail because the brand had no data; they fail because the creative pushed the claim past the data.

The cinematography enrichment matters here for a different reason than supplements. Skincare before/after needs to look honest, which means matched setup. Tonic's enrichment turns "before/after of a 30-year-old woman using serum X for 8 weeks" into a multi-element shot brief that locks lighting, lens, room, and posture across both sides. The output is structurally compliant by default; you have to actively fight it to produce a banned-shape ad.

For brands also working in regulated supplement categories, the same compliance principles apply but the ruleset differs. Don't assume the ASA's skincare expectations carry over to supplements one-to-one.

FAQ

Is it legal to generate a before/after video with AI for skincare?

Yes, with two conditions: the result has to be substantiated by evidence of the same level you'd need for a real before/after, and you have to disclose the AI generation. ASA's position is that synthetic content needs to be flagged so it doesn't mislead the audience.

What if the AI before/after is illustrative rather than a real result?

You still need substantiation for any claim implied by the comparison. "Illustrative" doesn't lower the bar; if anything, it raises the disclosure obligation. The asset should be labelled clearly as illustration, not styled as real customer evidence.

Can I use a real customer's quote alongside an AI-generated before/after?

This is a high-risk pattern. Pairing a real testimonial with synthetic visuals can create a misleading composite where the claim level is borrowed from the real customer but the proof is fictional. Either keep them separate or be very explicit about which is which.

What timelines does the ASA accept?

There isn't a hard rule, but realistic ones aligned to the actual mechanism of the product. Most cosmetic actives need 8-12 weeks to show meaningful change. Anything claiming dramatic results in days will get scrutinised even if the underlying data is sound.

Does AI disclosure protect me from a complaint?

Disclosure is necessary but not sufficient. You still need substantiation for claims. The AI label tells the audience the visual is synthetic; it doesn't tell them the claim is true. Both have to hold up.

---

100 free credits to test the skincare vertical's ASA pre-flight: tonicstudio.ai/signup?promo=UGC100.