AI lip sync re-syncs a person’s mouth movements in footage you already have to a new audio track. You need it when the video shows a real, identifiable person, when you are dubbing an ad you already shot, or when the read runs past 15 seconds. For new video built from blank, native-audio models generate synced speech in one step, so the separate pass is wasted work.
Lip sync looks like a quality question. You pick the tool with the cleanest mouth movement, process your clip, and ship. That view was right two years ago. It is wrong now, because the models that generate video also generate the speech inside it.
The question worth answering is narrower. Given the ad you are trying to make, is the separate lip sync pass real work or wasted work? The answer depends on whether a real person appears in the ad.
Key Takeaways
-
The dividing line is likeness, not quality. If a real, identifiable person appears in the ad, you need a lip sync or dubbing tool. If the person is synthetic or non-specific, native-audio generation does the job in one step.
-
Sora 2 refused the job by design. OpenAI’s video guide stated that real people, including public figures, could not be generated, and that character uploads depicting human likeness were blocked by default (developers.openai.com, September 2026). You could not prompt around that boundary. OpenAI removed Sora 2 and Sora 2 Pro from its API on 24 September 2026 (OpenAI API deprecations, accessed September 2026).
-
Veo 3.1 caps at 8 seconds. Google’s Gemini API documentation lists durations of 4, 6, or 8 seconds, with 8 required for 1080p and 4K (ai.google.dev, September 2026). A 30-second presenter read does not fit in one generation.
-
Dubbing is the strongest B2B case, because it is the same read run again for each market. YouTube reported averaging more than 6 million daily viewers watching at least 10 minutes of auto-dubbed content in December 2025 (blog.youtube, February 2026).
-
Disclosure rules now apply in New York and the EU. New York has required conspicuous disclosure of synthetic performers in ads since 9 June 2026, with a $1,000 penalty for a first violation. Article 50 of the EU AI Act has applied since 2 August 2026, and it asks deployers to label deep fakes.
-
DesignerBox runs video models that make speech in the same run. It does not lip sync footage you already shot.
What AI lip sync does
AI lip sync takes video you already have and re-animates the speaker’s mouth to match a different audio track. The original footage, framing, lighting, and camera movement stay. Only the mouth changes.
Lip sync is the last step of a dub. How AI dubbing works covers the steps before it.
That is a different operation from generating a video. A generator starts from a prompt or an image and produces new frames. A lip sync tool starts from your frames and edits them. The distinction sounds academic until you need your founder’s face in the ad, at which point it decides everything.
Four products get called “lip sync” and only two of them do this:
| Type | What it takes in | Example use |
|---|---|---|
| True video-to-video lip sync | Your footage plus an audio track | Re-cut a spokesperson read without a reshoot |
| Dubbing | Your footage plus a target language | Run one ad across six markets |
| Avatar generation | A script, plus an avatar you picked or trained | Presenter video with no shoot at all |
| Performance transfer | A driving performance plus a character reference | Act a character you generated |
Only the first two operate on footage you already own. Avatar tools generate a new clip and hand it back. If you needed the original preserved, they have not solved your problem.
Starting from a still image rather than video is a third job again, with its own model constraints. We cover that separately in how to make a photo talk with AI.
The dividing line is likeness, not quality
Whether a real person appears in the ad decides the tool, and the model vendors drew that line themselves. OpenAI’s API documentation for Sora 2 stated that real people, including public figures, could not be generated, and that character uploads depicting human likeness were blocked by default (developers.openai.com, September 2026).
Google applies a softer version of the same constraint. Veo 3.1 allows adults only in its image-to-video, first-and-last-frame and reference-image modes, and in the EU, UK, Switzerland and MENA “allow_adult” is the only setting (ai.google.dev, September 2026).
Read those together and the market splits cleanly. Native-audio generation owns new content featuring people who do not exist. Lip sync owns content featuring people who do. That is a consent and identity boundary the model vendors built on purpose, and no amount of output quality moves it.
So the decision runs on one question: does a real person appear in this ad.
What native-audio models can and cannot do
Several video models generate synchronized speech as part of the generation, with no second pass. Our guide to AI video generators with sound covers the three audio layers of an ad. The AI video prompting guide shows how each model wants the spoken line written. Verified specs, from each provider’s own documentation:
| Model | Native audio | Max duration | Source |
|---|---|---|---|
| Veo 3.1 | Yes | 4, 6, or 8 seconds | ai.google.dev, September 2026 |
| Seedance 2.0 | Yes | 4 to 15 seconds, multi-shot | docs.byteplus.com, September 2026 |
| Kling 3.0 | Yes | 3 to 15 seconds, multi-shot | kling.ai, September 2026 |
| Sora 2 Pro | Yes | Up to 20 seconds. Removed from OpenAI’s API on 24 September 2026 | developers.openai.com, September 2026 |
The duration column is the real constraint. Veo 3.1 stops at 8 seconds, which covers a hook and nothing longer. Sora 2 Pro generated clips of up to 20 seconds, and OpenAI removed it from the API on 24 September 2026. After that date, Seedance 2.0 and Kling 3.0 give the most room in this table, at 15 seconds. One model outside the table runs longer: ByteDance launched Seedance 2.5 on 31 July 2026, and its API generates 4 to 30 seconds at up to 1080p (seed.bytedance.com, accessed September 2026). For pipelines that ran on Sora, Sora alternatives by job names a replacement for each use.
Google is also candid about reliability on its own model page. It notes that creating videos with natural and consistent spoken audio, particularly for shorter speech segments, remains an area of active development (deepmind.google, September 2026). Budget for regeneration.
Google’s Gemini API lists Veo 3.1 audio as always on, and Google Cloud’s Veo 3.1 card lists sound generation as supported (docs.cloud.google.com, September 2026). Google’s deprecations page lists 22 October 2026 as the earliest shutdown date for the Veo 3.1 preview models in the Gemini API (Gemini API deprecations, October 2026). Specs in this category change often, so check the current docs before you commit a campaign to one.
When you still need a dedicated lip sync tool
You still need a dedicated AI lip sync tool in four cases. The footage already exists, a real person appears, you are localizing, or the read runs past 15 seconds.
The footage already exists. Native models generate. They do not retrofit. If you shot the ad last quarter and the offer changed, a generator cannot help you.
A real, identifiable person appears. Your founder, a paid actor, a licensed influencer. Sora 2 blocked this outright until OpenAI removed it from its API on 24 September 2026. Veo 3.1 constrains it.
You are localizing. The performance exists and needs to survive translation. This is the strongest commercial case in the category, and the volume data supports it. YouTube reported averaging more than 6 million daily viewers watching at least 10 minutes of auto-dubbed content in December 2025. Creators uploading multi-language audio saw over 25% of watch time come from the video’s non-primary language (blog.youtube, September 2025 and February 2026). Many ads need captions and new on-screen text before they need a dub, and what to change first when you localize a video ad sorts ads by what carries the message.
The read runs long. Veo 3.1 stops at 8 seconds in one generation, and Seedance 2.0 and Kling 3.0 stop at 15. A 30-second presenter read means joining several clips, or a lip sync pass on footage you filmed.
The reference case for all of this is still Cadbury’s Diwali campaign in India, which licensed Shah Rukh Khan’s likeness and used AI video technology to generate thousands of personalized retailer ads. It won a Titanium Lion at Cannes Lions in 2022 and the Creative Effectiveness Grand Prix in 2023 (contagious.com, September 2026). Note what made it work: a licensed likeness, used at scale, with consent at the center. For other campaigns and what AI did in each one, see the best AI commercials and how they were made.
Which tool does which job
The four products most teams shortlist differ on scope rather than on quality, and scope is what decides the buy.
| Tool | What it does with footage you upload | Scope note |
|---|---|---|
| ElevenLabs | Lip sync, in its Image and Video tools | Lip sync is not part of the Dubbing product and is available separately (help.elevenlabs.io, September 2026) |
| Synthesia | Lip sync through AI Dubbing | Framed around dubbing and translation rather than same-language re-cuts (synthesia.io, September 2026) |
| HeyGen | Video Translate and Lip Sync | Both work on an uploaded clip (heygen.com, September 2026) |
| Vozo | Lip sync, standalone or with dubbing | Sold as the re-sync step itself (vozo.ai, September 2026) |
Each vendor lists its plans on its own pricing page. Check that page on the day you buy, and read whether it shows the monthly or the annual rate before you compare two tools.
Avatar-first tools sit in a different bucket. Creatify, Argil and Tavus each train a reusable avatar from a video of a person, instead of re-syncing the original clip (creatify.ai, argil.ai and tavus.io, September 2026). Best fit if you want a repeatable presenter, outside scope if you need your existing footage preserved. For creator-style ads specifically, the best AI UGC ad tools compares that category on its own terms, and Arcads alternatives compares the AI actor tools.
The half DesignerBox covers
DesignerBox is AI creative production for brands and agencies. Scale your images, ads and video with AI and keep your brand on every piece. Build the workflow once with your brand rules, run it on every product, see the cost before each run, and keep everything from the first product photo to the finished ad in one place. It makes new video from your product and your brand profile, and a video model with native audio makes the speech in the same run. It does not include a lip sync tool that re-syncs footage you already shot. If that is the job, one of the tools above is the right buy. For the wider question of where the voice track itself should come from, AI voiceover for UGC ads compares the three production paths. A sung narrator is one more way to avoid lip sync, and how to make an advertising jingle with AI covers it.
Starting from a product still rather than a prompt is its own technique, and AI image to video for ecommerce covers which models hold a product through the motion. What DesignerBox covers is the other half of the split, and the half that repeats. You build the video ad once as a workflow, with your product, your brand rules and the model you pick for the video step. The workflow reads the brand rules on every run, so the next product runs the same way. When you publish the workflow as an app, a colleague completes a form and presses Run. The video ad templates are a place to start.
For a presenter clip with no footage at all, the Video with a person app makes the clip from an avatar and a script, with lip sync. DesignerBox vs HeyGen puts the avatar-first approach next to the campaign-first one, and HeyGen alternatives compares the tools most teams shortlist for this.
An 8-second clip costs 40 to 560 credits, depending on the model. AI video starts on the Premium plan. The free plan cannot make video. Plans and credits are on the pricing page. Our breakdown of AI video generation cost explains what moves the price of a clip.
Disclosure and consent
Lip-syncing a real person’s face is where this category meets the law, and the rules changed in 2026. This is general information, not legal advice.
New York. General Business Law section 396-b took effect on 9 June 2026. It requires anyone producing an advertisement to conspicuously disclose that a synthetic performer appears in it, where they have actual knowledge. Penalties are $1,000 for a first violation and $5,000 for each later one. The law does not apply to audio-only ads, or when AI is used only to translate a human performer’s words (nysenate.gov and governor.ny.gov, September 2026). The statute defines a synthetic performer as one who is not recognizable as any real, identifiable performer. Footage of a real, recognizable person falls under the consent rules below.
California. The state passed a similar law. The governor signed SB 1050 on 16 September 2026. It covers ads that prominently include a synthetic performer. Those ads need a clear disclosure, with wording close to “this performance features a synthetic performer”. The law takes effect on 1 January 2027, the default start date for a California law without an urgency clause (leginfo.legislature.ca.gov, accessed September 2026).
The EU. Article 50 of the EU AI Act has applied since 2 August 2026. If you publish a deep fake, you must say that it is AI-generated or AI-edited, at the latest when people first see it. A deep fake is realistic AI image, audio or video that could pass as real, and lip-synced footage of a real person can count as one. The machine-readable marking duty sits on the tool maker. The visible label sits on the deployer (European Commission, September 2026). The Commission’s guidelines, which are not binding, say that a brand that only hires an agency, and does not control how the agency uses AI, is not the deployer (Commission guidelines, September 2026).
Right of publicity. California Civil Code section 3344 names voice explicitly and requires prior consent for advertising use. Damages are $750 or the actual loss, whichever is greater, plus profits and legal fees (leginfo.legislature.ca.gov, September 2026). From 1 January 2027 the same law names digital replicas: California signed SB 1111 on 30 September 2026, and it says a person’s voice or likeness includes a digital replica (California SB 1111, accessed October 2026). New York Civil Rights Law section 50 requires written consent before you use a person’s name, picture, likeness or voice in advertising, and a breach is a misdemeanor (nysenate.gov, September 2026).
Platforms differ, and the direction surprises people. TikTok asks for disclosure on ads that contain fully AI-generated or heavily AI-edited images, video or audio. For non-Spark ads, you turn on the Ads Manager toggle. Spark ads follow TikTok’s Community Guidelines (ads.tiktok.com, September 2026). TikTok’s ads policy names voice cloning, and showing someone doing something they did not do, as edits that need the label. If TikTok finds undisclosed AI content in an ad, it rejects or restricts the ad (TikTok ads policy, September 2026).
Meta requires advertisers to disclose AI themselves in ads about social issues, elections or politics. Its ad policy says it began automated detection of third-party AI in ads on 1 June 2026, with no advertiser action needed (transparency.meta.com, September 2026). When Meta finds signals such as C2PA metadata, it adds an “AI info” label under About this ad, and Meta says this may not be available in every region. Meta also requires people to disclose organic posts with photorealistic AI video or realistic AI audio. So an organic AI video on Meta carries a disclosure duty that the same video running as an ordinary ad does not.
The practical standard for consent comes from the 2025 SAG-AFTRA Commercials Contracts, and it is worth following even outside union work. Consent must be clear and conspicuous, and based on a reasonably specific description of the intended use. The performer gives it in a Digital Replica Rider or in a separate signed writing (dglaw.com, January 2026).
Next steps
If a real person is in the frame, buy a dubbing or lip sync tool and skip the generators entirely. Once you have picked the approach, the order of the steps decides how real the result looks. Why AI talking video reads as fake covers the four tells and the fix.
If the ad is a short vertical spot rather than a presenter read, how to create TikTok video ads with AI covers the build. If the ad is new video built from your product, open the Short product videos template, add your product photo, and run it. See the templates.
FAQ
Can I lip sync a video with AI for free?
Some lip sync tools list a free plan. Before you use one for a paid ad, read what that plan allows for commercial use and whether it adds a watermark. Check the vendor’s live pricing page on the day you buy, and read whether the number on screen is the monthly or the annual rate.
Do I still need lip sync if I use Veo 3.1 or Seedance 2.0?
Not for new content. Both generate synchronized speech as part of the generation. You need a separate lip sync tool when the footage already exists, when a real identifiable person appears, when you are dubbing into another language, or when the read runs past 15 seconds.
Why can’t an AI video model make a video of my founder?
Vendors block or limit real faces on purpose. OpenAI’s video guide stated that real people, including public figures, could not be generated with Sora 2. OpenAI removed Sora 2 from its API on 24 September 2026 (OpenAI API deprecations, September 2026). ByteDance says Seedance 2.0 does not accept reference images or videos that contain real human faces, except through its consent and face check (docs.byteplus.com, September 2026). For a real person on camera, you need footage of them, their written consent, and a lip sync or dubbing tool.
Does DesignerBox do lip sync?
Not on footage you already shot. DesignerBox has a talking avatar with lip sync, which makes a new presenter clip from an avatar. It also runs video models that make speech in the same run, such as Veo 3.1 and Seedance 2.0. To re-sync mouth movements in existing footage, a dedicated dubbing or lip sync tool is the right choice.
Do I have to disclose AI lip sync in a paid ad?
The answer changes with the platform and the place the ad runs. TikTok asks for disclosure on ads with fully AI-generated or heavily AI-edited images, video or audio. Meta asks advertisers to disclose only in ads about social issues, elections or politics. New York has required conspicuous disclosure of synthetic performers since 9 June 2026. Article 50 of the EU AI Act has required deployers to label deep fakes since 2 August 2026.
What consent do I need to lip sync a real person?
Written consent, obtained before use, covering voice as well as likeness, and describing the intended use specifically. California requires prior consent for advertising use, and New York requires written consent. The 2025 SAG-AFTRA Commercials Contracts ask for consent that is clear and conspicuous and based on a reasonably specific description of the use, which makes a safe template.
Is dubbing the same as lip sync?
No. Dubbing replaces the audio track. Lip sync also re-animates the speaker’s mouth so it matches the new audio. Some tools do both, some only dub. ElevenLabs’ documentation, for instance, states that lip sync is not part of its Dubbing product and lives in a separate tool (help.elevenlabs.io, September 2026).
Sources
All accessed September 2026 unless a different date is given.
- Veo 3.1 native audio, clip durations, resolution requirements and person-generation limits: (ai.google.dev and deepmind.google, September 2026)
- Veo 3.1 sound generation on Google Cloud: (docs.cloud.google.com, September 2026)
- Sora 2 and Sora 2 Pro limits on real people and clip lengths: (developers.openai.com, September 2026)
- Sora 2 and Sora 2 Pro removal from OpenAI’s API on 24 September 2026: (developers.openai.com/api/docs/deprecations, September 2026)
- Seedance 2.0 durations and real-face rules: (docs.byteplus.com, September 2026)
- Seedance 2.5 launch on 31 July 2026, 4 to 30 seconds at up to 1080p: seed.bytedance.com, accessed September 2026
- Kling 3.0 durations, audio and multi-shot: (kling.ai, September 2026)
- YouTube auto dubbing and multi-language audio figures: (blog.youtube, September 2025 and February 2026)
- Cadbury “Shah Rukh Khan-My-Ad” awards: (contagious.com, September 2026)
- ElevenLabs Dubbing scope, and confirmation that lip sync sits in a separate product: (help.elevenlabs.io, September 2026)
- Synthesia’s uploaded-footage lip sync scope, framed around dubbing and translation: (synthesia.io, September 2026)
- HeyGen Video Translate and Lip Sync: (heygen.com, September 2026)
- Vozo lip sync: (vozo.ai, September 2026)
- Creatify, Argil and Tavus custom avatars from video: (creatify.ai, argil.ai and tavus.io, September 2026)
- New York synthetic performer disclosure law, effective 9 June 2026, and its penalties: (nysenate.gov and governor.ny.gov, September 2026)
- California SB 1050, synthetic performers in advertising, approved by the Governor on 16 September 2026: leginfo.legislature.ca.gov, accessed September 2026
- New York Civil Rights Law section 50 and California Civil Code section 3344: (nysenate.gov and leginfo.legislature.ca.gov, September 2026)
- California SB 1111, digital replicas named in Civil Code section 3344 from 1 January 2027, signed 30 September 2026: leginfo.legislature.ca.gov, accessed 2 October 2026
- EU AI Act Article 50 obligations and the Commission’s guidelines: (digital-strategy.ec.europa.eu and ec.europa.eu, September 2026)
- Meta ad disclosure rules and automatic AI labeling: (transparency.meta.com and facebook.com/business/help, September 2026)
- TikTok AI-generated content disclosure for ads: (ads.tiktok.com, September 2026)
- SAG-AFTRA 2025 Commercials Contracts, digital replica consent: (dglaw.com, January 2026)
- DesignerBox plans, credits and feature gating: DesignerBox pricing page (designerbox.ai/pricing), September 2026
Model capabilities verified against provider documentation from Google, OpenAI, ByteDance and Kuaishou as of September 2026. Tool scope verified from vendor pages as of September 2026. Legal and platform policy verified from primary sources including nysenate.gov, governor.ny.gov, leginfo.legislature.ca.gov, the European Commission, transparency.meta.com and ads.tiktok.com as of September 2026. California SB 1111 was read on 2 October 2026. This is general information, not legal advice. Model specs and platform policy in this category change monthly. Verify before you commit a campaign.
