You can make a talking photo with AI in about six steps: get the person’s written permission (or use your own face), choose a clear front-facing photo, write a short script, pick the voice, animate the face, then check the lip sync and label the result as AI.
A talking-photo or video tool animates the face to the words, and you should label the result as AI when you post it. Consent comes first because the face and the voice belong to a real person, and in many places that person has rights over both.
- Your own face, or written permission. A photo you are allowed to keep is not permission to make someone speak. Ask, and get the yes in writing.
- Never children, never without consent, never to deceive. Major tools and platforms ban all three, and so does the law in more and more places.
- Two kinds of tool: avatar tools that animate a photo to a script or recording, and video models that turn a photo into a clip with speech.
- Short script, clear photo. A front-facing face with the mouth visible and a line or two of plain speech give the best lip sync.
- Label it. YouTube, TikTok and Meta all ask for a label on realistic AI video of people.
A talking photo is a still picture of a face that an AI tool animates so the mouth, eyes and head move as if the person were speaking. Lip sync is the matching of those mouth movements to a voice track. The voice can be a real recording, a synthetic voice, or a clone of someone’s voice. If the face and the voice both belong to a real person, the result is a kind of deepfake, however kind the intention.
Before you start: whose face, and whose voice?
The plain point: you need permission for the face and, separately, for the voice. Your own face and voice are the simple case. Anyone else needs to say yes, in writing, to this specific use.
The reason is legal as well as polite. Control over your own face and voice is usually a right of publicity question, not a copyright one, so owning the photo does not give you the right to make the person speak. The US position, state by state and with dates, is on voice and likeness law. In the EU, the AI Act requires people who publish a deepfake to say it is artificial. 1
The tools say the same in their own rules. HeyGen, a talking-avatar tool, bans avatars of real people without their consent. 2 ElevenLabs, a voice tool, bans cloning a living person’s voice or likeness without consent, and recordings of anyone under 18. 3
A good permission covers five things. The reasons for each are on voice and likeness law:
- In writing. A message or email is a record; a nod is not.
- The face and the voice, named separately if both are theirs.
- What it is for and where it will be posted.
- For how long, and what happens if they change their mind.
- Whether a voice may be cloned (trained on their recordings), which is a bigger step than using one recording. See voice cloning.
Children’s faces or voices, a stranger’s photo, a celebrity, a politician, or anyone saying something they did not agree to say. Fake endorsements, fake apologies, fake confessions and anything used to trick someone out of money are the uses that make deepfake headlines and court cases. TikTok, for example, lists using “the likeness of private figures without consent” among content it does not allow. 4 How these fakes are used against people, and how to protect yourself, is on deepfake defence.
Consent is two permissions: one for the face, one for the voice. Get both in writing before you open a tool.
What you need
The plain point: a good photo, a short script, a voice, a tool, and the permission above.
- A photo: one face, looking at or near the camera, mouth visible and closed or slightly open, even light, sharp focus, with some space around the head and shoulders. Sunglasses, hands over the mouth and strong shadows make the result worse.
- A script: one or two short sentences to start. Plain words are easier to lip-sync than names and jargon.
- A voice: your own recording, the person’s own recording (with their consent), or a stock synthetic voice that belongs to nobody.
- A tool of one of the two kinds below.
Reasoning, 12 Oct 2026 — the photo checklist is this site’s own advice for clearer lip sync, not a vendor requirement; check your tool’s own upload guidance.
The two kinds of tool
The plain point: avatar tools are built for a face talking to camera; general video models can do more, but give you less control over the exact words and timing. These are examples, not recommendations.
- Talking-photo and avatar tools. You upload a photo and a script or a recording; the tool animates the face to match. HeyGen is one example. Its developer documentation is clear that having no consent step in the software is not the same as having permission. 5
- Video models that generate speech. You give the photo as the first frame and write the spoken line into the prompt. xAI says its Grok Imagine Video 1.5 adds lip-synced speech, and Google’s Veo docs differ: the Gemini API page asks for spoken words in quotation marks, while Google Cloud’s newer guide says to write speech after a colon and avoid quotation marks, so the words are not drawn on screen.6 7 8 Google’s Gemini app, which can turn photos into videos, adds its own rule. 9
Prompting speech in video models, layer by layer, is on sound and dialogue in AI video. Turning a photo into a clip without speech is on how to turn a photo into a video with AI.
The steps
The plain point: permission, photo, script, voice, generate, check, label.
- Get permission in writing using the template above, or use your own face and voice.
- Choose and prepare the photo. Crop it so the face is large and centred. Do not “improve” the face with a beauty filter first; it drifts further from the real person.
- Write the script short. Read it aloud with a timer. If it takes longer than the clip, cut words, not pauses.
- Choose the voice. A real recording gives the most natural result. A synthetic voice needs direction: pace, pauses and which word carries the line (see prompting music and voice).
- Generate. In an avatar tool, upload the photo and the script or audio. In a video model, attach the photo as the first frame and use a prompt like the one below.
- Check it closely (next section). Regenerate or trim anything wrong.
- Show the person before you post, if it is not you. Their yes was to the idea; let them see the result.
- Label it and post it using the platform’s AI label (see below).
A clear photo, a short script and a locked camera do more for lip sync than any setting. Keep everything except the mouth still.
How to check the result
The plain point: watch it with the sound on, then with the sound off, then frame by frame around the hardest words.
- Mouth shapes: do the lips close on “b”, “m” and “p”? Missing closures are among the easiest lip-sync errors to spot.
- Teeth and tongue: do they smear or change shape between frames?
- Eyes: natural blinks, not a fixed stare or constant blinking.
- The words: listen for wrong or slurred words, especially names. The model may say something close to your script rather than your script.
- Identity: does it still look like the person at the end of the clip? Put the first and last frames side by side.
- Timing: does the mouth start when the sound starts?
Reasoning, 12 Oct 2026 — this checklist is this site’s own; it describes what to look at, not a measured error rate.
Labelling: what YouTube, TikTok and Meta ask for
The plain point: all three ask you to label realistic AI-made or AI-altered video of people, and a talking photo of a real person is exactly that. Use the platform’s own label when you upload, and say it in the caption too.
| Platform | What it asks | In its own words |
|---|---|---|
| YouTube | Disclose realistic AI content using the “AI use” setting in YouTube Studio when you upload. Making a real person appear to say something they did not is named as needing disclosure. | 10 |
| TikTok | Label AI or edited content that realistically shows people or scenes, with TikTok’s AI-generated label or a clear caption or sticker. | 4 11 |
| Facebook, Instagram, Threads (Meta) | Use Meta’s disclosure tool for photorealistic video or realistic-sounding audio that was made or altered digitally. Meta shows an “AI info” label. | 12 13 |
One detail from YouTube’s page is useful for creators: cloning your own voice for a voiceover or dub is listed among the edits that do not need disclosure. 10 A talking photo of your own face is still realistic AI video of a person, so label that. Some tools also add their own marks; the background is on watermarking and content credentials.
If a viewer could believe a real person said it, label it. All three big platforms ask for that, and the label costs you nothing.
Questions people ask
Is it legal to make a talking photo of someone?
With their clear consent and an honest label, it is usually low risk; without consent it can break publicity, privacy or deepfake laws, depending on where you and your viewers are. The US picture is a state-by-state patchwork, set out with dates on voice and likeness law, and in the EU deepfakes must be disclosed. 1 This is not legal advice.
Can I animate a photo of a relative who has died?
Many people do this privately as a keepsake. Ask the close family first, keep it private unless they agree otherwise, and know that in several US states publicity rights continue after death and are held by the estate, as voice and likeness law explains. Reasoning, 12 Oct 2026 — this site’s own advice; the legal point about rights after death is sourced on the linked page.
Do I have to label a talking photo?
On YouTube, TikTok and Meta’s apps, yes, if it looks realistic. TikTok’s rule is one sentence. 14
Can I use a photo of a celebrity?
Not without their permission. Fame makes a likeness more valuable to protect, not less, and tools such as HeyGen ban avatars of public figures without explicit consent. 2
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. According to the page, does owning a photo of someone give you the right to make them speak?
- Yes, if you took the photo yourself
- Yes, as long as you label it as AI
- No, you need their permission, ideally in writing
Show the answer
Answer: No, you need their permission, ideally in writing. The page says a face and voice are usually a right of publicity question, so owning the photo is not permission, and the yes should be in writing.
2. What does the page say to do before posting a realistic talking photo on YouTube, TikTok or Meta’s apps?
- Use the platform’s AI label and say so in the caption
- Nothing, as long as the person agreed to it
- Add a filter so it looks less real
Show the answer
Answer: Use the platform’s AI label and say so in the caption. The page says all three platforms ask you to label realistic AI video of people, so use the platform’s own label and say it in the caption too.
3. What does the page suggest for the clearest lip sync?
- A long script full of names and technical terms
- A clear front-facing photo, a short plain script and a still camera
- A beauty-filtered photo so the face looks its best
Show the answer
Answer: A clear front-facing photo, a short plain script and a still camera. The page says a clear photo, a short script and a locked camera do more for lip sync than any setting, and that a beauty filter makes the face drift.
Related on this site: what a deepfake is · voice cloning · text-to-speech · how to spot fake media · sound and dialogue in AI video · What Is Lip Sync in AI?
- 1EU AI Act, Article 50, via artificialintelligenceact.eu, read at source 12 Oct 2026deployers of an AI system “that generates or manipulates image, audio or video content constituting a deep fake” “shall disclose that the content has been artificially generated or manipulated.” The page lists Article 50 as coming into force on 2 August 2026, and narrows the duty for “evidently artistic, creative, satirical, fictional or analogous” work. ↑aa deployer of a deep fake system “shall disclose that the content has been artificially generated or manipulated.” ↑b
- 2HeyGen, Content Moderation Policy, effective 25 Jul 2024, read at source 12 Oct 2026“Unless explicit consent has been provided, creating avatars of other individuals is strictly prohibited.” and prohibited avatars include those that “Represent real individuals, including celebrities or public figures, without their explicit consent”. ↑aprohibited avatars include those that “Represent real individuals, including celebrities or public figures, without their explicit consent”. ↑b
- 3ElevenLabs, Prohibited Use Policy, last updated 9 Oct 2026, read at source 12 Oct 2026“Cloning, synthesizing, or impersonating the voice, identity, or likeness of any living individual” “without their explicit and informed consent or other valid authorization”, and “You may not submit voice recordings of anyone under 18 to create or train a voice model.” ↑
- 4TikTok, Community Guidelines: Integrity and Authenticity, released 25 Aug 2026, effective 24 Sep 2026, read at source 12 Oct 2026not allowed includes “Using the likeness of private figures without consent” and “AI-created likenesses made to bully or harass”. ↑a“We require clear labeling when AI or editing is used to realistically depict people or scenes.” ↑b
- 5HeyGen, Avatar consent, developer docs, read at source 12 Oct 2026“Having no API step to complete is not permission to use someone’s likeness.” ↑
- 6Google Cloud, Best practices for generating videos, Gemini Enterprise Agent Platform documentation (the older Vertex AI address redirects here), last updated 9 Oct 2026, read at source 13 Oct 2026“To prevent the model from rendering text in the video, use a colon ( : ) after the speaker’s action to denote speech and avoid using quotation marks”; its recommended example is “A woman says: My name is Clara.” ↑
- 7xAI, Video generation, read at source 12 Oct 2026image-to-video “Generates video with the provided image as the starting frame.” and “Generated videos include an audio track by default, with lip-synced speech on `grok-imagine-video-1.5`”. ↑
- 8Google AI for Developers, Generate videos with Veo 3.1 in Gemini API, read at source 12 Oct 2026“Dialogue: Use quotes for specific speech.” ↑
- 9Google, Gemini Apps Help, Generate videos with Gemini Apps, read at source 12 Oct 2026“Only upload photos you have the right to use.” and “For now, this feature isn’t available to users under 18.” ↑
- 10YouTube Help, Disclosing use of GenAI content, read at source 12 Oct 2026“We require creators to disclose AI-generated or altered content that seems realistic.” (wording re-read at source 12 Nov 2026) Disclosure is needed for content that “Makes a real person appear to say or do something they didn’t do.” Creators who repeatedly do not disclose may face “penalties from YouTube, including removal of content or suspension from the YouTube Partner Program.” ↑aexamples that do not need disclosure include “Cloning one’s own voice to create voice overs or dubs.” ↑b
- 11TikTok Newsroom, New labels for disclosing AI-generated content, 19 Sep 2023, read at source 12 Oct 2026the label can be used on “any content that has been completely generated or significantly edited by AI.” ↑
- 12Meta, Labeling AI-Generated Images on Facebook, Instagram and Threads, 6 Feb 2024, updated 1 Apr 2025, read at source 12 Oct 2026“We’ll require people to use this disclosure and label tool when they post organic content with a photorealistic video or realistic-sounding audio that was digitally created or altered, and we may apply penalties if they fail to do so.” ↑
- 13Meta, Our Approach to Labeling AI-Generated Content and Manipulated Media, 5 Apr 2024, read at source 12 Oct 2026“we’re updating the “Made with AI” label to “AI info” across our apps”. ↑
- 14TikTok, Community Guidelines: Integrity and Authenticity, effective 24 Sep 2026, read at source 12 Oct 2026“We require clear labeling when AI or editing is used to realistically depict people or scenes.” ↑
EU AI Act, Article 50, via artificialintelligenceact.eu, read at source 12 Oct 2026.
HeyGen, Content Moderation Policy, effective 25 Jul 2024, and Avatar consent (developer docs), both read at source 12 Oct 2026.
xAI, Video generation; Google AI for Developers, Generate videos with Veo 3.1 in Gemini API; Google, Generate videos with Gemini Apps; all read at source 12 Oct 2026.
TikTok, Community Guidelines: Integrity and Authenticity (released 25 Aug 2026, effective 24 Sep 2026) and TikTok Newsroom, New labels for disclosing AI-generated content (19 Sep 2023), read at source 12 Oct 2026. TikTok’s help-centre page on AI-generated content loads only with JavaScript and could not be read by this site’s fetcher on 12 Oct 2026.
Meta, Labeling AI-Generated Images on Facebook, Instagram and Threads (6 Feb 2024, updated 1 Apr 2025) and Our Approach to Labeling AI-Generated Content and Manipulated Media (5 Apr 2024), read at source 12 Oct 2026.
The permission template, prompts, photo and checking lists are this site’s own, not claims from these sources. Tools are named as examples only, with no affiliate links. Nothing here is legal advice.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK