AI UGC video generator: product ads with lip-synced AI actors
OpenShorts' AI Shorts turns a product URL or a short description into a vertical UGC-style ad: Gemini researches the product and writes 20 to 25 second scripts (hook, problem, solution, demo, call to action), an AI actor says it with lip-synced voice, two b-roll inserts cover the problem and the product, and the captions are burned in. Output is a 1080x1920 MP4 for TikTok, Reels and Shorts.
The video generation runs on your own fal.ai and ElevenLabs keys: about $0.65 a video in Low Cost mode and about $2 in Premium, billed by those providers. On OpenShorts Cloud the scripts come out of your plan's minutes, the free plan included; self-hosted, OpenShorts is free under MIT and you add a Gemini key.
Open AI Shorts in the app and add your fal.ai and ElevenLabs keys: about $0.65 a video billed by them. The script runs on your plan, free plan included; paid plans from $12/month.
What AI UGC is, and what this generator makes
UGC, user-generated content, is the style of ad that looks like a person talking to their phone camera about something they use, rather than a produced commercial. It works on TikTok and Reels because it looks like the posts around it. The AI Shorts tab makes that format without a person or a camera: an AI-generated actor delivers a script written for your product, with real lip movement, cut with b-roll and captioned the way short-form video is.
Every video follows the same five-beat structure, because that is the shape a 20-second product ad has to have:
| Beat | Time | On screen |
|---|---|---|
| Hook | 0 to 5 s | The actor, talking to camera |
| Problem | 5 to 9 s | B-roll of the pain point, voice over it |
| Solution | 9 to 16 s | The actor presents the product |
| Demo | 16 to 21 s | B-roll of the product in use |
| Call to action | 21 to 25 s | The actor, pointing to the link in bio |
The result is a vertical 9:16 MP4 at 1080x1920, H.264 with AAC audio, with captions burned in a few words at a time, the way short-form viewers expect to read them. Each script also comes with a suggested caption for the post.
Put another way, it is a UGC AI video generator for one format only: the short vertical ad. The AI UGC scripts are written first and shown as text, so you choose the angle before anything is generated or paid for.
How to make an AI UGC video, in three steps
1. Describe the product or paste its website
Open the AI Shorts tab in the app. Paste the product's URL (the page and up to 3 of its subpages are read, and the product is researched on the web) or describe the business in your own words. Pick the language, the actor's gender, a script style and how many scripts you want.
2. Choose a script, an actor and a voice
Each script is a 20 to 25 second ad in five beats: hook, problem, solution, demo, call to action. Pick one, edit the narration if you like, then choose an actor (from the gallery, from three new AI portraits, or your own photo) and a voice from your ElevenLabs account.
3. Generate, then download or post
Choose Low Cost (about $0.65 a video) or Premium (about $2), billed by fal.ai and ElevenLabs on your keys. You get a 1080x1920 MP4 with the lip-synced actor, two b-roll inserts and burned-in captions, ready to download or post to TikTok, Instagram and YouTube.
What you put in
- A website URL, a description, or both. With a URL, the page and up to 3 subpages are read and Gemini researches the product on the web, then extracts the value proposition, the audience, the pain points and the angles worth trying. With only a description (up to 10,000 characters), the scripts are written from what you say.
- Language. English or Spanish, written natively in each rather than translated.
- Actor. A woman or a man.
- Style. UGC natural (authentic, talking to camera), educational, shock or discovery (a surprising opener), storytelling (a mini narrative arc) or before and after.
- How many scripts. 1, 2, 3 or 5 per run (the API accepts up to 10). Each script takes a different pain point or angle, which is the point: you test angles, not one ad.
It is not limited to software. The product analysis works from whatever the page or the description says, so a restaurant, a shop, an online course, a coaching practice or a local service gets scripts about its own customers' problems.
The AI actor and the voice
After you pick a script you choose who says it. There are three ways:
- The actor gallery: portraits generated for earlier videos, ready to reuse so a series of ads keeps the same face.
- Three new portraits: describe the person or keep the script's suggestion, and three options are generated (about $0.06 for the three).
- Your own photo: upload a picture and that face is animated and lip-synced. Use only your own face or someone who has agreed to it.
The voice is any voice in your ElevenLabs account, previewable before you generate, including voices you created there. The narration is shown as editable text, so you can fix a product name or change a line before anything is paid for. Speech is generated with ElevenLabs' multilingual model, and the captions are then transcribed from that audio so they match what is said word for word.
Low Cost or Premium
The two modes share the script, the voice, the b-roll and the captions; they differ in how the talking head is made. The estimates below are the ones the app shows before you generate, and the providers' own prices decide the final bill.
| Step | Low Cost | Premium |
|---|---|---|
| Actor image (Flux 2 Pro) | $0.05 | $0.05 |
| Voice (ElevenLabs) | about $0.10 | about $0.10 |
| Talking head | Hailuo 2.3 Fast image-to-video, $0.19, then VEED Lipsync, $0.20 | Kling Avatar v2 from image and audio, about $1.69 |
| Two b-roll inserts | about $0.10 | about $0.70 |
| Compositing and captions | $0 (FFmpeg) | $0 (FFmpeg) |
| Per video | about $0.65 | about $2 to $2.50 |
Low Cost is the one to test angles with: three Low Cost videos cost about what one Premium video does. Premium generates the face and the mouth together, so head movement and expression look more natural, and it is worth it for the script that won the test. The b-roll in both modes is an AI still image animated with a slow zoom and pan, which reads as motion in a feed without the price of generated video. A Premium talking head takes a few minutes to render; failed steps can be retried and reuse the parts already generated, so a failure late in the pipeline does not pay for the image and the voice twice.
What it costs, and who bills you
There are two bills, and keeping them apart is the honest way to describe it.
The video generation (actor image, talking head, lip-sync, b-roll, voice) is billed by fal.ai and ElevenLabs, to the keys you connect in Settings. That is true on OpenShorts Cloud and self-hosted alike: no plan includes it yet. The keys are kept in your browser and sent with each request.
The research, the scripts and the posting are OpenShorts. On Cloud they run on the managed Gemini and posting keys and each analysis run uses 1 minute of your plan's balance, which the free plan (your first video up to 60 minutes, then 20 minutes a month, watermarked clips, no card) covers too. Paid plans, from the pricing table:
| Plan | Price | Minutes a month |
|---|---|---|
| Cloud Free | $0, no card | First video up to 60 min, then 20 min. Clips watermarked |
| Cloud Starter | $12/month | 100, no watermark |
| Cloud Creator | $29/month | 300, no watermark |
| Cloud Pro | $59/month | 750, no watermark |
| Self-hosted | $0, MIT | No meter; your hardware and your own API keys |
Self-hosted, OpenShorts is free under MIT with no meter; you add your own Gemini key for the research and scripts, plus the same fal.ai and ElevenLabs keys.
AI UGC or a human creator
Hiring a UGC creator is usually quoted somewhere between $50 and $500 a video, depending on the creator and the usage rights. An AI UGC video here costs about $0.65 to $2 in generation, takes minutes instead of a brief, a shipment and a wait, and can be made in five variants to see which angle works before spending more.
What the AI version cannot give you is the thing that makes UGC persuasive in the first place: a real person who really used the product. The actor has never held it, the b-roll is illustrative rather than footage of your product, and viewers who look closely can tell. The sensible split is to use AI UGC to find the hook and the angle cheaply, then put a real customer or creator on the script that won, and to never present an AI actor as a real customer.
Use cases
- Testing ad angles. Five scripts from five pain points, generated in Low Cost, posted or run as ads; keep the one that holds attention.
- Apps and software. The original use: a URL in, ads about the problems the product solves.
- E-commerce products. A product page becomes a hook, a problem and a reason to tap the link.
- Restaurants and local businesses. A description of the place and what it is known for, in English or Spanish.
- Coaches and course creators. Talking-head promos for a programme, without filming another one.
- Agencies. First drafts of UGC concepts for a client, before a creator is booked.
Vertical formats, and where to post them
Every video is 9:16 at 1080x1920, the native shape of TikTok, Instagram Reels and YouTube Shorts, and about 20 to 25 seconds, well inside the limits of all three. From the result screen you can post to TikTok, Instagram and YouTube at once or schedule it for a date and time. TikTok receives it as a draft, which you finish in the app; that is also where you add a trending sound, which posting through an API cannot. If you already have long videos of your own, the clip generator is the other half: it cuts real footage into Shorts and TikToks, while AI Shorts makes videos when there is no footage at all. For the thumbnail of the long video itself, see the AI thumbnail maker.
Disclosure: saying it is AI
An AI actor is a synthetic person, and the platforms and advertising rules treat that as something viewers may need to know. TikTok, YouTube and Instagram each have their own label for realistic AI-generated content, with their own rules for when it is required, and advertising rules in many countries forbid presenting a paid or invented endorsement as a genuine customer review. Those rules change and differ by market, so check the current ones for where you post. OpenShorts does not add a label or a mark to the video; whether and how to disclose is your decision, and labelling it is rarely a disadvantage with an audience that can usually tell anyway.
Rights and what you are responsible for
- Faces. Generated actors are not real people. An uploaded photo must be you or someone who consented to being animated in an ad.
- Claims. The script is written from your page and from web research. Read it before generating: you are the advertiser, and the claims in the ad are yours.
- Providers' terms. The image, video and voice come from fal.ai's models and ElevenLabs, under your accounts and their terms, including what they allow for commercial use.
- The public gallery is opt-in. A checkbox lets you share a finished video in the OpenShorts gallery; when ticked, the video, the product name and the script are public. It is off unless you tick it.
Compared with avatar video subscriptions
Most AI avatar video tools sell a monthly subscription with a quota of videos or minutes. SendShort, for example, includes AI avatar videos from its Professional plan, $29/month, or $23/month billed yearly (checked 5 October 2026). OpenShorts works the other way round: the code is open source, there is no avatar quota, and you pay the generation providers per video on your own keys, so a month with two ads costs two ads. The trade-off is setup: two provider accounts and keys instead of one subscription, and a narrower product, built only for short vertical product ads.
What it does not do
- It does not make long videos. Scripts are 20 to 25 seconds, by design.
- It does not cover the generation cost on any plan yet. fal.ai and ElevenLabs bill you directly.
- It does not write in languages other than English and Spanish.
- It does not add background music. Add a sound when you finish the post, as TikTok drafts invite you to.
- It does not draw a separate hook headline on screen. The hook is spoken in the first seconds and appears in the captions.
- It does not film your product. The b-roll is AI imagery of the problem and the product category, not your real interface or packaging.
- It does not train a persistent avatar. Each video starts from one image; reuse a gallery actor to keep the same face across a series.
Common questions
- How much does an AI UGC video cost?
- About $0.65 a video in Low Cost mode and about $2 in Premium (the in-app estimate shows up to $2.50 for Premium), paid to fal.ai and ElevenLabs on your own keys. On OpenShorts Cloud the script and orchestration come out of your plan's minutes (the free plan included); plans run from $0 to $59/month. Self-hosted, OpenShorts itself is free under MIT and you also bring a Gemini key.
- Is there a free AI UGC video generator?
- Not in the sense of free videos. On OpenShorts Cloud the research and the scripts run on the free plan, from its minutes, so you can get UGC scripts for nothing; the video itself (actor, lip-sync, voice, b-roll) is always billed by fal.ai and ElevenLabs on your own keys, about $0.65 a video in Low Cost mode. Paid Cloud plans start at $12/month; self-hosted, OpenShorts is free under MIT and you pay only those providers and your own Gemini key.
- Do I need my own API keys?
- Yes, for the video. Generation always uses your fal.ai key (actor image, talking-head video, lip-sync, b-roll) and your ElevenLabs key (voice), on Cloud and self-hosted alike, and those providers bill you directly. On Cloud the Gemini part (research and scripts) and posting run on the managed keys; self-hosted, you supply a Gemini key too.
- What do I need to give it?
- Either the product's website URL or a written description of up to 10,000 characters, or both (the description then adds context). With a URL, the page and up to 3 subpages are read and the product is researched on the web before the scripts are written.
- How long are the videos?
- About 20 to 25 seconds, by design. Every script follows a fixed five-part structure (hook, problem, solution, demo, call to action), which is the length and shape of a typical short-form product ad.
- What format is the video?
- A vertical 9:16 MP4 at 1080x1920, H.264 video with AAC audio, captions burned in. It fits TikTok, Instagram Reels and YouTube Shorts without any reframing.
- Can I use my own face or a real person as the actor?
- Yes: upload a photo and it is animated and lip-synced to the voiceover. Only use a photo of yourself or of someone who has agreed to appear in your ads; animating a real person without consent can break the law and the platforms' rules.
- Which languages are supported?
- English and Spanish. Scripts are written natively in the chosen language (casual US English, or Spanish), and the voice comes from ElevenLabs' multilingual model, so it is not a translation of an English script.
- What is the difference between Low Cost and Premium?
- Low Cost animates the actor image with MiniMax Hailuo 2.3 Fast and then lip-syncs it with VEED Lipsync, for about $0.65 a video. Premium generates the talking head with Kling Avatar v2 from the image and the audio in one step, for about $2, with more natural movement. Both use the same script, voice, b-roll and captions.
- Is it an AI avatar video generator?
- In the sense that it makes talking-head videos of an AI-generated person, yes. It is built for one job, short vertical product ads, rather than long presenter videos or training content, and it does not train a persistent avatar of you: each video is generated from a single image.
- Can I post the videos straight to social media?
- Yes, to TikTok, Instagram and YouTube, now or scheduled, from the result screen. TikTok receives it as a draft you finish in the TikTok app, where you can add a trending sound. Each script also comes with a suggested caption.
- Do I have to label it as AI?
- Rules differ by platform and by country, and they change. TikTok, YouTube and Instagram all have their own labels for realistic AI-generated content, and advertising rules generally forbid presenting an actor as a real customer. Check the current rules of the platform you post on; OpenShorts does not add a label for you.
Sources
- Pipeline, models and per-step estimates:
saasshorts.pyanddashboard/src/components/SaaShortsTab.jsx; limits incloud/config.py. Source: github.com/mutonby/openshorts. - Provider prices are the estimates the app shows, 29 October 2026; fal.ai and ElevenLabs set the final price on their own pricing pages.
- SendShort plans: our comparison data, checked 5 October 2026. See SendShort compared.
- OpenShorts plan prices and minutes: the pricing table.