Three seconds. That is the entire pitch.
Magic Hour AI, the Oakland video platform founded by David Hu and Runbo Li, announced this week that its AI UGC ad generator is now free to try: pick an actor, paste a script, and an open-weight video model called LTX-2.3 renders a synthetic talking head pitching your product. No signup, no credit card, no creator outreach. You get three generations a day, each three seconds long, at 480p. The whole thing runs in your browser.
Three seconds sounds like a joke until you remember how UGC ads work. Nobody watches the whole thirty seconds. The entire economics of the format live in the first three: the hook. Get it wrong and the scroll keeps scrolling. Get it right and you have bought another twenty-seven seconds of attention. Magic Hour just made the hook free.
This is not really a story about a free tool. It is a story about what happens when the most expensive part of a creative format collapses into its most disposable unit.
What it is
On October 2, Magic Hour AI announced a free tier for its AI UGC Ad Generator, powered by LTX-2.3, Lightricks’ open-weight video model. The workflow is deliberately simple: choose or upload an actor image, set the opening shot (pose, background, the product that appears in the first frame), paste your script or upload audio, and generate. Renders take three to ten minutes, and you can queue several at once.
The full tool ships with more than 90 AI actors, or you can upload your own creator image. Magic Hour’s pitch is blunt: no creators, no revision cycles, scroll-stopping ads in minutes. Its own FAQ admits the honest version: AI generation beats creator coordination when you need to test lots of scripts fast, while hiring a real creator still wins when you need a specific person, a real location, or hands-on product demonstration.
The free tier is the hook factory: three generations a day, three seconds each, 480p, no account needed. The catches are printed in normal type, which is refreshing. Free outputs can carry a watermark, free users are limited to personal non-commercial use, and commercial rights arrive with a paid subscription or credit pack. The LTX-2.3 model behind it is open weights under Lightricks’ community license, which permits commercial use for companies under $10M in annual recurring revenue.
LTX-2.3 is not some toy checkpoint. Released by Lightricks in March 2026, it is a 22-billion-parameter audio-video model with native vertical video support, synchronized dialogue and sound in a single pass, and it currently ranks first among open-weight video models on the Artificial Analysis Text-to-Video Arena. The engine got good enough that someone built a UGC ad factory on top of it and gave the first three seconds away.
What changed
The UGC ad economy was built on a supply chain: brief a creator, negotiate a rate, wait for the footage, request revisions, edit, ship. A single creator video could take a week and a few hundred dollars. The synthetic version takes a script and ten minutes.
That supply chain collapse is the headline, but the pricing is the real story. The free tier is engineered around the one unit of creative that matters most in paid social. Three seconds is not a limitation, it is the product. You cannot make an ad with three seconds, but you can make a hook, and three free hooks a day turns hook testing from a budget line into a reflex.
Magic Hour is explicit about where synthetic UGC wins: e-commerce, SaaS, digital services, supplements, beauty, fashion, online courses. Testimonials, product demos, and problem-solution narratives, the formats with the highest engagement. In other words, the formats where the script matters more than the soul behind it.
And the timing is spicy. The FTC’s 2023 Endorsement Guides already treat a computer-generated character reading a paid script as an endorser: virtual does not mean exempt. FTC staff guidance from March 2025 adds that AI-generated endorsements must be clearly disclosed, and that a labeled fake testimonial is still a fake testimonial. Meanwhile the EU’s AI Act asks deployers to label realistic synthetic media at first exposure. The legal scaffolding already exists. The free tooling just caught up.
What works
Use it as a hook laboratory, because that is what it is. The winning workflow is not “fire your creators.” It is: generate ten hooks by lunch, spend real money only on the two that survive, then shoot those with a real human holding the real product.
This is where the smartest performance teams will live. Synthetic actors are perfect for volume: openers, angles, tones, languages, objection handling, retargeting variants. Every variation costs minutes, not days. The tool even lets avatars physically hold products in the first frame, which covers the demo-adjacent cases where the product is the star and the person is furniture.
There is also a localization angle hiding in the feature list: script changes and multiple languages mean one concept can be re-skinned for different markets without reshoots. And do not sleep on the boring superpower: no revision cycles. The synthetic actor never gets tired, never ghosts you, never posts the raw footage to its own account by mistake.
What breaks or stays fenced
Plenty, and the fences are load-bearing.
First, the free tier is a tasting menu, not a meal. Three seconds, 480p, watermark, no commercial rights. You cannot run a campaign on it. It is an onboarding funnel for the paid product, which is fine, but do not confuse “free” with “free lunch.”
Second, the trust problem is the whole ballgame. UGC worked because it borrowed credibility from real people with real faces and real audiences. A synthetic actor has no audience and no credibility. It is a spokesperson, not social proof. The format looks like UGC but functions like a spokesperson ad wearing a UGC costume. Marketers who price synthetic creative like it carries creator trust will learn the difference in their CPA.
Third, the law is already there and it bites. A synthetic testimonial that implies a real customer experience is deceptive with or without an AI label, and the FTC’s fake-review rules now carry civil penalties. Labels must appear before the consumer engages, in the first line or the first frames. Three-second hook ads make that interesting, because the entire asset is the opening seconds.
Fourth, the platform layer is moving. TikTok, YouTube, and Meta all have synthetic-media labeling toggles, and realistic unlabeled synthetic faces are exactly what those systems are built to catch.
Fifth, everyone gets the same actors. Ninety-plus faces sounds like a lot until ten thousand brands are testing hooks with the same woman in the same kitchen. Creative sameness is the silent killer of synthetic UGC. Winners will customize: their own actor images, their own products in the first frame, their own scripts with actual teeth.
Who it is for
- Performance marketers running paid social at volume. If your bottleneck is creative iteration speed, this is a hook-testing machine that costs nothing to try.
- DTC and e-commerce teams in the categories Magic Hour names: beauty, supplements, fashion, gadgets. If your best format is the direct-to-camera problem-solution pitch, the synthetic version gets you ten variants before lunch.
- Agencies that need concepts fast. Storyboard-grade creative you can actually watch beats a deck every time.
- Anyone localizing creative across markets. One concept, five scripts, five languages, zero reshoots.
It is not for brands whose entire edge is authenticity. If your customers buy because they trust the person holding the product, a synthetic face is a downgrade, not a shortcut.
What to do this week
- Run the free tier as a hook test. Pick your best-performing UGC ad, write three new hooks for it, generate all three before lunch. You just ran a creative test that used to cost a week and a creator fee.
- Separate the spokesperson from the proof. Map your creative: which assets borrow credibility from a real person, and which just need a face and a script? Synthetic actors are for the second pile only.
- Write your synthetic disclosure policy now. Decide where the AI label lives (first frames, first line of copy) before you scale anything. The FTC’s trigger is how the content functions in front of an audience. If a reasonable viewer would think the presenter is real, disclose.
- Customize or die of sameness. Upload your own actor image, put your actual product in the first frame, write scripts with teeth. The default library is a commodity. The customization layer is the brand.
- Keep your real creators. The winning stack is synthetic volume plus human trust: AI for the fifty hooks, humans for the five winners and everything that needs a real face, a real kitchen, a real result. The creator economy does not die here. It gets promoted to the work that cannot be faked.
Sources
- Magic Hour AI announcement: “Magic Hour launches free AI UGC ad generation with LTX-2.3,” ABNewswire, October 3, 2026 (datelined October 2, 2026)
- Magic Hour AI UGC Ad Generator product page and FAQ (magichour.ai/products/ai-ugc-ad-generator), accessed October 3, 2026
- “Lightricks Releases LTX-2.3, Open-Source Video Model With Sharper Detail and Native Vertical Format,” VP Land, March 2026; Wikipedia, LTX-2
- “Lightricks LTX-2.3 Showcases Theatrical Voice Synthesis in Open-Weight Audio-Video Generation,” The Agent Times, April 2026
- “FTC endorsement plus an AI label,” Versely Studio, September 2026 (FTC Endorsement Guides, fake-review rule, EU AI Act Article 50)
- “AI-Generated UGC Regulations and FTC Disclosure Compliance: Guide Released,” Marketers Media (GetHookd), May 2026
- “FTC AI Content Disclosure Rules: What Brands Must Know in 2026,” Human Ads AI (March 2025 FTC staff guidance)




