Best AI lip sync and talking photo tools, and what you must have before you animate anyone

Lip sync is the AI trick where a mouth moves to match audio. It is useful for greetings, translated versions of a message and explainer videos. It is also the easiest way to hurt someone, because it puts words in a real person's mouth. So this page puts consent first, then tools, then quality.

Limits and prices below come from each vendor's own pages, read on 2 November 2026, and from Higgsfield's pages and terms. Check each vendor's live page before you pay, because plans change.

One of the tools I mention, Higgsfield, pays Blur Studio a commission if you subscribe through my link. It does not change what I say about consent, which applies to every tool here equally.

This page may contain affiliate links from which Blur Studio earns commissions.

Consent first: what you must have before you animate anyone

A photo you can see online is not a photo you may animate. Animating a real person's face or cloning their voice is a use of their likeness, and the rules are stricter than for ordinary pictures. Treat the list below as the minimum:

1. Written release from the person in the photo. It should name AI animation and the specific uses (website, ads, social). A general photo release may not cover it. See the photo and media release form.
2. Separate written consent for their voice, if you use their real voice or a clone. Consent to be filmed is not consent to be cloned. See AI voice cloning tools and consent.
3. No lookalikes of real people, and no celebrities or public figures, unless you have a signed licence from them.
4. No minors. Do not animate children, and do not make avatars that appear to be under 18.
5. Keep the records in the project folder, and honour takedown requests quickly.
6. Never use it for fake testimonials. A talking photo of a "customer" who never said those words is a fabricated review. Allowed uses are things like a greeting from the real person, a narrated explainer beside a real demo, or a translated version of a message the real person approved (see putting the CEO in a localised ad for the sign-off steps).

What the tools themselves say:

• HeyGen, an AI avatar video platform, has a moderation policy (effective 25 July 2024) that requires the explicit consent of the individual for custom avatars, prohibits avatars of real people such as celebrities or public figures without explicit consent, and prohibits avatars that appear to represent people under 18.
• Higgsfield's terms require you to have the rights and consents for any name, likeness or voice in what you upload, ask you not to submit biometric templates (stored maps of a face or voice), and say not to use the service to identify or authenticate people. Higgsfield's privacy policy says biometric data from face or voice features is not stored and is destroyed after processing; read the current policy for yourself.
• Magic Hour, a vendor of talking photo tools, says on its own blog that paid subscriptions and credit packs give watermark-free video and commercial use under its service terms, but you still need permission to use the image and the voice.

I would not rely on any claim that a "photo avatar" is consent-free. If a page tells you consent records are unnecessary, check it against the written policies above. The law on likeness is moving. Laws on synthetic media and personal likeness are changing in many places, and I am not going to describe bills I could not confirm. Check your local law and the platform's current rules before you publish. This is information, not legal advice.

Disclosure matters too. Where law or a platform requires it, say plainly that the presenter is synthetic. Higgsfield's terms say you must disclose AI where required by law and must not remove provenance markers (hidden labels that show a file was made with AI).

Two jobs, two kinds of tool

• Talking photo: you have only a still image plus audio. The tool animates the face. Examples: D-ID, Magic Hour's talking photo tool, and Higgsfield's Intone (a talking-head tool in its model list).
• Lip sync for video: you already have video of a person speaking and want the mouth to match new audio, for example a corrected line or a translation. Examples: Sync, or the video option in Higgsfield's Lipsync Studio, whose page says you can add "any photo or clip" as the base.

Choose by input. If you have video, do not rebuild it from a still, because you lose the original performance and the camera work. Many roundups mix the two jobs together, which is why their rankings do not help.

Quick picks

• A low-cost way to try a talking photo: D-ID's trial or Lite plan, both personal-use only (check the live page, since Magic Hour also offers a short free test).
• Existing video with new audio: Sync.
• Repeatable presenter videos and translation, with an explicit written consent policy: HeyGen.
• Lip sync as one part of a wider image and video plan: Higgsfield.

The options, with limits and costs from each vendor page

D-ID

D-ID makes talking presenter clips from a photo. Its pricing page (annual-billing rates shown) lists a Trial at $0 for 14 days with 3 minutes and a full-screen watermark, Lite at $4.7 a month ($56 billed yearly) with 10 minutes a month, a D-ID watermark and a personal-use licence, Pro at $16 a month ($191 yearly) with 15 minutes, an AI watermark and a commercial-use licence, and Advanced at $108 a month ($1,293 yearly) with 100 minutes.

• Best for: a simple presenter clip on a tight budget.
• Skip if: you need a watermark-free file on the cheapest plan, since the commercial licence starts at Pro.

Sync (sync.so)

Sync changes the mouth in a video so it matches new audio. Its pricing page shows plans of Hobbyist at $5 a month plus usage (videos up to 1 minute, 1 job at a time), Creator at $19 a month plus usage (up to 5 minutes, no watermark), Growth at $49 and Scale at $249. Usage rates at base prices: its lipsync-2 model at $0.05 per second, its lipsync-2-pro model at $0.08325 per second, and higher for its newer models. A dubbing rate of $0.10 per second of audio was listed, with a note that a change was rolling out.

• Best for: existing footage, fixed lines and dubbing where you want to keep the original performance.
• Skip if: you only have a still photo.

HeyGen

HeyGen makes videos with AI avatars that speak your script, and it can translate videos. Its pricing page, read on 2 November 2026, lists a Free plan (3 videos a month, watermarked, up to 1 minute) and Creator at $29 a month ($24 a month billed yearly); check the page for current plans. Its written consent policy is explicit about real people, which is a reason to look at it first for that work. I did not compare it line by line with every other vendor's policy, so read each one.

• Best for: repeatable presenter videos and translation at volume. See ElevenLabs vs HeyGen vs Synthesia.
• Skip if: you want a one-off clip from a single photo.

Magic Hour

Magic Hour is an AI video site with a talking photo tool. Its own blog says the free version allows 3 guest attempts a day with no signup, a 5 second duration and one animated face per photo, and the output may carry a watermark. Paid plans give watermark-free video and commercial use. That is a vendor statement, so test it. Its listicles rank Magic Hour first, so I do not use them to rank tools.

Higgsfield

Higgsfield's own pages describe a Lipsync Studio that uses several lip sync models, including its own Speak 2.0 and models from other companies, and a tool called Intone, which Higgsfield has described as making a talking-head video from one image and a short voice recording (check the tool for its current input limits). Lipsync Studio's own page says you start by adding any photo or clip, then audio you upload or type, then pick a model. The price table lists lip sync clips from about 2 credits for a 2 second Kling Speak clip (Kling is a video model) up to about 58 credits for an 8 second clip from Veo 3 with audio (Veo 3 is Google's video model). A credit is the unit the site spends per generation, and clip lengths differ by model. For example, Higgsfield Speak 2.0 at 720p is listed at about 14 credits for 5 seconds. Check the live pricing page for current plans and costs, because they change.

I found nothing showing that Higgsfield is better than a specialist for any single lip sync job. Its Terms do not restrict commercial use of outputs, but its pricing table marks commercial use as not included on the Free plan, so use a paid plan for client work. Its value is that lip sync sits beside image and video generation in one plan, which helps if you are making a whole short video. For a one-off talking photo, D-ID is simpler, and for dubbing existing footage, Sync is built for it. For the details, see Higgsfield lipsync and Higgsfield pricing.

• Best for: creators who also need image and video generation in the same place.
• Skip if: lip sync is the only job, or you need a clear published per-second price.

Other routes you will see on other lists

Two more routes come up often. I read about both in Magic Hour's lip sync roundup on 2 November 2026, which is a vendor page, so check the original sources before you rely on them.

• Performance transfer (Runway Act-Two). Instead of matching a mouth to new audio, you film yourself acting a line and a character image copies your speech, expression and motion. Magic Hour's roundup says Runway retired its separate Lip Sync tool on 27 May 2026 and points users to Act-Two, with clips up to 30 seconds. Choose this when the performance, not only the mouth, should drive the character. Read Runway's current page for price and limits.
• Open source on your own computer (LatentSync). LatentSync is a research model from ByteDance that takes an existing video and new audio. It costs no subscription, but it needs a capable graphics card and technical setup, and you must check its licence before using output for clients. It suits a developer, not a shop owner. Related older tools such as Wav2Lip carry similar trade-offs.

To be fair to those pages: Magic Hour's roundup also tells readers to test the same clip across tools and record cost per accepted clip, so that advice is not unique to this guide. What this guide adds is the consent gate before any test and the written rights chain for client work.

Free options: what they really give you

• Trials usually mean a short length, a visible watermark, or no download. D-ID's trial is 14 days, 3 minutes and a full-screen watermark.
• Magic Hour's free talking photo is a few attempts, short and may carry a watermark.
• Higgsfield's free tier adds a visible watermark to every generation, and it does not include credits. Help center text says there is no trial period.
• Some free plans generate but block the download. Always run the free test, export a file and look at it before you pay for a year.

How to test quality: same photo, same audio, score sheet

Use identical inputs for every tool. One sharp, front-facing portrait you have permission to use, and one 20 second audio file that includes a name, a number, a hard sentence with p, b, m and f sounds, and a pause. Score each result from 1 to 5:

Check	What to look for
Mouth closure	Lips close on p, b and m
Teeth	No smear or extra teeth
Blinks and eyes	Natural, not frozen
Head drift	No stretching at the neck, no jumps
Identity	The face stays the same person in every frame
Silence	The mouth is still when there is no sound
Last word	Not cut off

Pass for a tool: every score is at least 3 and no frame changes the person's identity. Look hard for failure modes: glasses and beards that warp, side-on faces that fail, fast speech that slips, and singing, which needs its own test.

Costs per accepted clip

Work it out with a table. Using the vendor figures above: D-ID's Pro plan gives 15 minutes a month, so about 30 clips of 30 seconds if none are rejected. Sync's lipsync-2 at $0.05 per second makes a 30 second clip about $1.50 in usage, on top of the plan. Rejected clips count too. A realistic budget divides the plan price by the clips you actually accept.

Billing traps that VEED's lip sync guide passes on from user reviews of other tools (VEED sells video tools and ranks itself first, so treat these as leads to check, not proof about those companies): Trustpilot reviewers cited credits that do not roll over for Hedra, the guide says the Jogg AI free plan lets you generate but not download, one Trustpilot reviewer of Captions said many avatars were removed two months after an annual purchase, and free plans often add watermarks. Run the free test and check the cancel path before you commit.

Rights chain for a paid client job

1. Photo: a licence or signed release.
2. Voice: your own recording, a licensed text-to-speech voice (a computer voice that reads your script), or a clone made with written consent.
3. Tool: a plan that allows commercial use and removes the watermark you do not want.
4. Disclosure line if the presenter is synthetic and law or platform rules ask for it.
5. A record of all of the above in the project folder.

A 7 day test with pass and fail marks

1. Day 1. Collect releases. Pick one portrait and one script.
2. Day 2 and 3. Run the same inputs through three tools on free tiers.
3. Day 4. Score with the sheet above and pick one.
4. Day 5. Make the real clip on the cheapest plan that is watermark-free and allows commercial use.
5. Day 6. Internal review plus one outside viewer: "does anything feel off?"
6. Day 7. Publish with a disclosure line if needed. Then cancel or keep.

Pass: the outside viewer cannot name a visual flaw on the first watch, and cost per accepted clip is under your target. Fail: otherwise, or if any release is missing.

Who should skip lip sync

• You cannot get written consent from the person in the picture.
• The plan is a testimonial or endorsement from someone who did not say it.
• The video will go to people who would feel tricked to learn the speaker is synthetic.
• You only need a voiceover. Skip the face and use a voice tool. See AI voiceover and dubbing tools.
• You can simply record a real 30 second video on a phone.

How I'd approach this

I would get the signed release first, then write the script, then run the same-inputs test on two or three tools before paying for any plan. I would pick a specialist tool for a single job, and only use a broader plan such as Higgsfield's multi-model subscription when lip sync is one piece of a bigger video. Every published clip gets a disclosure line if there is any doubt, and the consent paperwork goes into the project folder. For a related guide, read how to make a talking presenter video without filming.

If you do this often, Blur Studio can build the repeatable workflow around the tools: a consent form that is signed online, a shared asset library and an approval page. See how I work with businesses.