← Back to blog

How to Make a Before-and-After Ad with AI

How to make a before-and-after ad with AI: multi-reference conditioning for state consistency, Meta's 2026 policy limits, honest-claims rules, and prompts.

You can make a before-and-after ad with AI in four steps: generate a consistent "before" state, hold the same face, room, and lighting for the "after" using multi-reference conditioning, choose split-screen or sequential structure, and export vertical. A finished before-and-after ad runs about $5 to $12 in model credits versus the $150 to $600 a creator charges. Before-and-after is one of the highest-CTR ecommerce formats, but it is also one of the most policy-restricted, so this guide covers both the production and the platform rules that decide whether your ad runs at all.

TL;DR

Why before-and-after converts (and why it's fragile)

The format works because it shows the outcome the buyer wants in a single glance. The contrast does the persuading, no voiceover required. That is also why it is fragile: the entire effect depends on the viewer believing the two states are the same subject at two points in time. If the face subtly changes, the room shifts, or the lighting warms up in the "after," the brain reads a swap rather than a transformation and the ad dies. Consistency is not a nice-to-have here; it is the format.

The 4-step workflow

Step 1: Generate and lock the "before"

Build the "before" state and lock every element you intend to carry into the "after." Generate the base frame in Nano Banana Pro, and if a person is on camera, lock their identity in Higgsfield Soul 2.0 so the same face appears in both states.

Portrait of a woman at a bathroom mirror, dull uneven skin tone, soft morning window light from the left, plain tiled background, front-facing, photorealistic, neutral expression

Save this frame. It is the reference that every "after" element must match: face, room, camera angle, and light direction.

Step 2: Generate the "after" with multi-reference conditioning

This is the step that keeps the format honest-looking. Upload the "before" frame as a reference into Seedance 2.0, whose multi-reference conditioning changes only what you prompt while holding the rest constant.

Same woman, same bathroom mirror, same morning window light from the left, same tiled background and camera angle, now with visibly clearer, more even skin tone, front-facing, photorealistic

The "same X" repetitions are load-bearing. They tell the model to preserve room, angle, and light so only the result changes. For the on-camera face, keep the Higgsfield lock so the person is unmistakably the same across both states. Generate three variants and select the pair where nothing but the result shifts.

Step 3: Split-screen or sequential cut

Two structures, different jobs.

Split-screen: before and after side by side (or stacked for 9:16) at the same time. Best when the contrast is the entire message and you want it readable in under a second. See what a split-screen is for the layout mechanics.

Sequential cut: before holds for 1 to 2 seconds, then a transition reveals the after. Best when you want the reveal to feel like a payoff and to give the viewer a beat of anticipation. For paid social, the sequential cut usually holds attention longer because it builds a small arc.

Step 4: Export, disclose, and check policy

Export 9:16, add captions, and apply the AI-generated content label. Then, before you spend a dollar, check whether your vertical can even run a before-and-after on your target platform, because this is where many of these ads get rejected.

Platform policy: what Meta allows in 2026

Meta's rules on before-and-after imagery are specific, and they were updated in 2026. As of the July 2026 policy update, Meta enforces on a claims basis rather than blanket product bans, but the restrictions that matter for this format are:

The practical read: skincare-glow, makeup, hair styling, home-and-room, and general cosmetic transformations are workable within claim limits; weight loss and anti-aging before-and-afters are not a viable Meta format. TikTok and other platforms have their own health-and-wellness restrictions, so verify each one for your category before running. Policies change; confirm the current version before a campaign.

Honest claims

Beyond platform policy, truth-in-advertising rules apply. An AI-generated before-and-after depicts an idealized result, so it must represent a result the product can actually deliver, and it must not present the avatar as a real customer's documented outcome (that crosses into the fake-testimonial territory the FTC prohibits). Frame it as a disclosed illustrative demonstration, keep the depicted change within what your product substantiates, and avoid date stamps or "in 7 days" overlays unless you can back the timeline. The disclosure label does not exempt you from the honesty of the claim.

Cost math versus a creator

Line item AI before-and-after Creator before-and-after
Before/after stills (Nano Banana Pro) $0.16 included
After-state motion (Seedance 2.0, 3 variants) $1.80 included
On-camera face lock (Higgsfield Soul 2.0) $3 to $8 included
Total per finished cut $5 to $12 $150 to $600
Turnaround same day 5 to 14 days

A creator before-and-after that requires filming a genuine transformation is slow and hard to control. AI lets you test the visual and the framing cheaply, so long as your vertical is one where the format is permitted. Where it is not (weight loss, anti-aging on Meta), the cheapest production in the world does not matter, so check policy first.

FAQ

Can I run a weight-loss before-and-after ad on Meta?

No. As of the 2026 policy, Meta prohibits before-and-after transformation comparisons for weight-loss products, including implied transformations like a product shown next to a fit body. Anti-aging and wrinkle-treatment before-and-afters are also prohibited. General cosmetic before-and-afters are allowed if the ad makes no violative claim. Verify the current policy before your campaign, since Meta updates these rules.

How do I keep the face and room identical between before and after?

Use multi-reference conditioning in Seedance 2.0: upload your locked "before" frame and repeat the constant elements in the prompt ("same room, same light, same angle") so only the result changes. For an on-camera person, lock their identity in Higgsfield Soul 2.0 so the same face appears in both states. Generate several variants and pick the pair where nothing but the outcome shifts.

Split-screen or sequential, which converts better?

It depends on the message. Split-screen is best when the contrast is the whole pitch and you want it readable instantly. Sequential (before holds, then reveals the after) usually holds attention longer on paid social because it builds a small anticipation arc. Test both; they are cheap to produce, and the winner varies by product and audience.

Do I need to disclose an AI before-and-after?

Yes on paid placements: TikTok and Meta require the AI-generated content label at upload. Beyond the label, an AI before-and-after must depict a result the product can actually deliver and must not be presented as a real customer's documented outcome. Keep it framed as an illustrative demonstration and keep the depicted change within what you can substantiate.


The workflow is four steps and $5 to $12 in compute: lock the before, change only the result with multi-reference conditioning, choose split-screen or sequential, and export. The harder gate is policy, so confirm your vertical is allowed before you produce.

Lock your before-state at app.8frame.co and generate the after against it. For the compliance-heavy sibling of this format, see how to make a testimonial-style video with AI, and for the full UGC toolkit, how to make a UGC ad with AI.

Related articles

use caseHow to Make a Day-in-the-Life Video with AIuse caseHow to Make a Founder Story Video with AIuse caseHow to Make a GRWM Video with AI (Get Ready With Me)

Make it
move.

Stay in the loop

Be the first to hear about our launch and get product updates