Lesson 5 of 7
Make a photo talk
7 min read
Give the tool a single photo of a person and a recording of a voice, and it can make that face say the words — mouth moving in time, as if they were filmed. How does a still photo start talking?
A face plus a voice
This is a different trick from the clips we've made so far. Instead of describing a scene, you give the tool two things: a still image of a face — a photo, or a generated portrait — and an audio track of speech. The tool animates the face so it appears to speak those exact words, with natural little head movements. The result is a talking avatar, and it's how a lot of AI presenters and explainer videos are made.
A still face plus a voice track, animated so the person appears to speak, is a talking avatar. You supply the face and the audio; the tool supplies the movement.
Matching mouth to sound
The heart of it is getting the mouth right. Every speech sound has a mouth shape — your lips press together for an m, round for an o, show teeth for an f. The tool listens to the audio, works out which sound is happening at each instant, and draws the matching mouth shape on that frame. Line those shapes up with the sounds and the face looks like it's really talking. That matching is called lip-sync.
Lip-sync is matching each mouth shape to the sound being spoken at that moment. Get the shapes and the timing right and a still face reads as genuinely talking.
Handy — and easy to misuse
Talking avatars are genuinely useful: a presenter who never needs a studio, an explainer recorded once and refreshed by editing the script, or the same person appearing to speak a language they don't. But the same trick makes a real person appear to say things they never said. That's the line between a helpful avatar and a deepfake.
Only make a talking avatar of someone with their clear permission — and, where it matters, say that it's AI-generated. Making a real person appear to say things they never said, without consent, is exactly how harmful deepfakes are made.
The gist
- —A talking avatar = a still face + a voice track, animated so the person appears to speak
- —Lip-sync is the core: each speech sound is matched to its mouth shape, frame by frame
- —Powerful for presenters and dubbing — but only with consent; without it, that's a deepfake
You feed a portrait photo and an audio recording into a talking-avatar tool. What is the tool mainly doing to make it convincing?
Continue in the app
Take the whole AI Video course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.