Face swap lets an ad team keep one winning script, hook and product shot and change only the person on screen. That makes it a cheap way to test which presenter, age group or local creator a market responds to. This guide covers which tools fit ad work, how to run a clean persona test, what each variant costs, and the consent and platform rules that decide whether the ad runs at all.
Morphed makes one of the four tools compared here. The comparison uses each tool's published product pages and pricing, checked on November 4, 2026, plus Morphed's own model catalog and credit prices. We did not run a side-by-side generation test, so this guide makes no claims about which tool's output looks best.
What are the best AI face swap tools for advertising campaigns?
The best AI face swap tools for ad campaigns are Magic Hour and Akool for swapping a face into footage you already shot, HeyGen for teams already producing avatar presenter videos, and Morphed for teams that generate their ads and want persona variants with a known credit cost. The deciding question is whether your test needs your original footage kept frame for frame, or whether a newly generated clip of the same scene is acceptable.
| Tool | Swaps a face in existing video | Published starting price | Usage rights for ads | Main con for ad work |
|---|---|---|---|---|
| Magic Hour | Yes (photo, video, GIF) | Creator $19/mo, or $12/mo billed yearly | Commercial use on all paid plans | Video swap is sold as an hours allowance per plan, not a per-second price |
| Akool | Yes (MP4, MOV, AVI, MKV, WEBM) | Starter $12/mo billed yearly | Pricing page lists a personal license below the Business plan | Business license starts at the Business plan, $174.30/mo billed yearly |
| HeyGen | Yes (MP4 and GIF, plus photos) | $29/mo, or $24/mo billed yearly | Commercial rights on paid plans | Built around avatars; free exports carry a watermark |
| Morphed | Yes, with H3 Max Recast (Pro plan); other routes generate a new clip | $9/mo Basic, 30% off the first 3 months | Commercial use on paid plans, with permission for every likeness | No free renders, and no frame-by-frame face tracker |
Sources: Magic Hour pricing, Magic Hour face swap, Akool pricing, Akool face swap, HeyGen face swap, HeyGen pricing, and Morphed's pricing.
Magic Hour is the most direct fit when you have a finished ad and want to drop a different face into it. Its pricing page includes commercial use and watermark-free exports on all three paid plans, and its face swap page asks users to get consent from anyone whose face they use. The con: the pricing page gives video face swap as a total number of hours per plan, so you work out cost per variant yourself.
Akool supports video and image face swap, swapping several faces in group photos, and up to 50 uploads at a time. Its face swap page lists face swap at "4+ credits", without an exact per-swap cost. The con for agencies is the license row on the pricing page: Starter, Pro and Pro Max show a personal license, and the business license starts at the Business plan.
HeyGen has a face swap tool for photos, MP4 and GIF, including multi-face swaps in one pass, and says the tool is built for consented, disclosed use. Paid plans add commercial rights, and API access is listed at $0.05 per second of video. The con: HeyGen's main pricing page does not mention face swap, so check the limits for your plan before you plan a large test.
Morphed suits teams that already make AI UGC ads or product videos with AI and want face variants in the same workspace. It offers three ways to change the person on screen, described in the next sections. The cons: there is no free tier, the photo swap is prompt-guided editing, so likeness varies between generations, and the route that keeps your original clip, H3 Max Recast, needs the Pro plan.
Which face swap tool works on videos for ad campaigns?
A face swap tool that works on video ads either replaces the face inside your original footage or generates a new clip of a new person doing the same thing. Pick based on what your test needs to keep identical.
- Your footage stays, only the face changes. Magic Hour, Akool and HeyGen's face swap tool work this way. In Morphed, H3 Max Recast is the closest equivalent: it swaps the people in a 5 to 30 second clip for up to 4 reference photos and keeps the motion, camera, cuts and sound.
- The performance stays, the clip is new. Morphed's Kling Motion Control takes a photo of your person and a reference video of up to 30 seconds and generates a new clip of that person doing the same movement. It can keep the original sound.
- A still frame becomes the ad. Swap the face on one frame, then animate it with image-to-video. This is the cheapest route, and it suits short hooks and 5 to 15 second clips.
Morphed has no frame-by-frame face tracker. Every video route produces a newly generated clip, so check each take for likeness drift in fast head turns and profile shots before it goes into a test.
How do you swap faces in marketing videos to test different personas?
To test personas with face swap, hold every part of the ad constant except the face, get written consent from each person, and run the variants side by side with equal budgets. A test where the script, voice or first frame also changes can't tell you whether the face made the difference.
- Lock one master ad. Use the same hook, script, product shots, captions and edit for every variant. If the voice changes with the persona, note it, because it is now a second variable.
- Cast with consent. Use creators or actors who signed a release that covers AI alteration, paid ad use, the platforms, the markets and how long the ad will run. Stock photos of strangers and public figures are out.
- Make the variants from the same source. Always edit from the original master frame or clip, not from the previous variant. In our 40-edit drift test, Nano Banana 2.1 Edit followed instructions best but visibly changed the rest of the image when edits were chained.
- Review every take. Reject any variant where the face flickers, the product or text changed, or the likeness drifted. A broken take tests the artifact, not the persona.
- Label and launch. Name each variant by persona, apply each platform's AI disclosure (see below), and run them in the same campaign with equal budgets until each has enough results to compare.
The same workflow in Morphed
- Photo-led ads: open the AI face swap tool, add the target photo and the face to use, and describe the swap. It runs in the image editor on a model such as Nano Banana 2.1 Edit. Swap one face per generation for the most reliable result.
- Short video hooks: swap the face on your first frame, then animate it in the video generator. This is the video face swap route most teams start with.
- One performance, many faces: in Motion Control, the Bulk cast tab takes 2 to 20 people photos and one reference video, and returns one video per person. That is a persona test set from a single upload.
- Keep the original clip: choose H3 Max Recast in Motion Control (Pro plan) to swap up to 4 people in a 5 to 30 second clip while keeping its cuts and sound.
How much does each face-swapped ad variant cost?
In Morphed, a face-swapped ad variant costs from 11.5 credits for a photo to 645 credits for a 15-second Recast of an existing clip. The table uses Morphed's live credit prices and the monthly credits of the Basic ($9, 650 credits) and Pro ($29, 2,000 credits) plans.
| Variant type | Credits per variant | Variants per Basic plan | Variants per Pro plan |
|---|---|---|---|
| Photo swap, Nano Banana 2.1 Edit at 1K | 11.5 | 56 | 173 |
| Photo swap + 10 s H3 Max Turbo clip (768P) | 71.5 | 9 | 27 |
| Photo swap + 10 s H3 Max clip (768P) | 141.5 | 4 | 14 |
| 15 s Kling v3 Motion Control re-performance | 270 | 2 | 7 |
| 15 s H3 Max Recast of an existing clip (768P) | 645 | Pro plan required | 3 |
| 15 s H3 Max Recast (1080P) | 967.5 | Pro plan required | 2 |
New subscribers get 30% off the first 3 months on web checkout, and one-off credit packs start at 300 credits for $6. Creating an account and planning with the agent are free, but every render uses credits.
The other three tools don't publish a per-second price for video face swap on their consumer plans, so a like-for-like cost per variant isn't possible from published data. The nearest figures: HeyGen lists API face swap at $0.05 per second of video, Magic Hour sells video face swap as hours per plan, and Akool lists face swap from 4 credits.
How do you localize a UGC ad for different markets with face swap?
To localize a UGC ad with face swap, keep the product, script structure and edit, and swap in a consenting creator who fits each market. Then translate and re-voice the script separately. A face swap alone doesn't change the language, so pair it with a localized voiceover, captions, and any on-screen prices or offers for that market.
Two rules matter more when an ad crosses borders. First, the creator's release has to name every market you plan to run in. Second, local law can add disclosure duties on top of platform rules. In the EU, Article 50 of the AI Act requires anyone deploying AI to make deepfakes to disclose that the content was artificially generated or manipulated, from 2 August 2026. For producing the base creative, see our guides to AI UGC video generators and making AI video ads.
How do you A/B test creator faces without breaking the test?
To A/B test creator faces, change only the face, keep the voice, script, first frame and budget identical, and judge on the metric the face can actually affect. The face mostly affects the first seconds, so watch hook and click metrics before conversion metrics, which depend on the landing page and offer too.
Keep each test small: two to four personas against one control is easier to read than ten variants splitting the same budget. Retire a variant once it clearly loses, rather than letting a weak face keep spending. And keep the consent records next to the creative files, because a winning variant is the one you will want to scale into new markets later.
Consent, likeness rights and ad platform rules
A face-swapped ad needs three things before it runs: the person's permission, the platform's disclosure, and a use that doesn't imply a false endorsement. A disclosure label does not fix missing consent.
Consent and likeness rights. Get written permission from every person whose face appears in the final ad. It should cover AI alteration, paid advertising, the platforms, the markets and the term. Never use a celebrity, public figure or a private person who hasn't agreed, even in an internal test that might leak. This is general guidance, not legal advice. Right-of-publicity rules differ by country and by US state, so check with counsel before a large campaign.
| Platform | What it requires for AI-altered people in ads | Official source |
|---|---|---|
| Meta | Social issue, election and political ads must disclose photorealistic images, video or audio that were digitally created or altered, including a real person shown doing or saying something they didn't, or a realistic person who doesn't exist. Ads that should disclose and don't are rejected. For other ads, Meta adds "AI info" to ads made with its own generative AI tools and, since June 2026, to ads it detects were made with third-party AI tools. When the image shows an AI-generated photorealistic person, the label can sit next to "Sponsored". | Meta political ads AI disclosure policy, how AI ads are labelled, Meta's June 2026 update |
| TikTok | Significantly AI-edited or AI-generated ad content is allowed only with the AIGC label or a clear disclaimer, caption, watermark or sticker. Showing the main subject doing something they didn't do counts as significant. Undisclosed AI content is rejected or restricted. Using a public figure's likeness without permission is prohibited, and ads using a private person's likeness without permission are removed. | TikTok misleading and false content ad policy, AI disclaimers in Ads Manager |
| Election ads with synthetic content that inauthentically depicts real or realistic-looking people must tick the "Altered or synthetic content" box, and some formats also need the advertiser's own visible disclosure. Separately, the Misrepresentation policy bans ads that impersonate a public figure or imply their endorsement to get people to pay money or share information, and accounts that break it are suspended. | Google synthetic content disclosure, Google Misrepresentation update |
Policies were checked on November 4, 2026 and change often, so confirm the current version in each ad platform's policy center before launch.
Morphed's policy. Morphed's Acceptable Use Policy requires consent from the person whenever a feature processes photos of real people. It bans images of public figures or celebrities without their explicit permission, impersonating real people without consent, and deceptive deepfakes. It also requires you to disclose AI-generated content where the law or context calls for it, and it doesn't allow Morphed content in political campaigns or elections. Prompts and outputs pass through automated safety filters, flagged content can go to human review, and violations can end in account suspension. Paid plans include commercial use, as long as you have permission for every likeness in the ad.
Start a persona test
For a first test, swap two or three consenting creators into one winning hook with the video face swap route, or start from stills with the AI face swap tool. To compare the wider field of consumer tools, see our ranking of the best face swap AI tools.