AI Talking Avatar: Make Any Character Speak

An AI talking avatar is a picture of a face that speaks a line with its lips in sync, and Flick makes one from a single image and a line of text or audio.

By Zoey Z.Last updated

An ink drawing of a portrait on an easel, its subject speaking

Most AI talking avatar tools animate a stock presenter reading a script, which suits a sales video but not a character in a story. In a film, the avatar has to keep its face, speak in its own voice and say the words you wrote and no others, take after take. We read the prompts behind 19 talking clips from 5 films made in Flick by the team and the Summer 2026 Filmmaker Residency, from a broadcast announcer on a CRT screen to an 18 year old arguing with his mom. This page shows the three ways Flick makes a character speak, four steps for a talking avatar with Character Dialogue, and the prompt habits that kept these lips in sync.

Make a talking avatar in FlickSee the three ways

Describe your avatar and its line

Describe the face, the line and how it should sound, or pick one of the real examples below, and it opens as the first request in a new Flick project.

What is an AI talking avatar?

An AI talking avatar turns a still picture of a character into a video of that character speaking, with the mouth, face and head moving to the words.
The three ways Flick makes a picture speak, by what you start with
You start withUse in FlickWhat you get
A picture and a recorded lineCharacter Dialogue, with Kling AI Avatar v2 or Kling AI Avatar v2 ProThe face in the picture speaks your audio, and keeps its appearance
A picture and a typed lineImage to Video, with a model that makes sound, such as Kling O3 Pro, Seedance 2.0 or MiniMax H3 MaxA shot that starts on your picture and speaks the quoted words in a voice the model makes
A picture and a voice clipOmni Reference, with MiniMax H3 or Seedance 2.0A new shot of the character saying new words in the voice from the clip
  • Character Dialogue keeps your exact recording, so use it when the line is already voiced, by you, an actor or Text to Voice.
  • Image to Video and Omni Reference also move the camera and the body, so use them when the avatar sits inside a scene.

How do you make a talking avatar in Flick?

Make a clear picture of the face, record the line, then give both to Character Dialogue, which moves the mouth and face to the sound of your audio.

  1. Step 1

    Make the face

    Generate or upload a still where the face is large, lit and turned toward the camera, because Character Dialogue animates the face it finds in that one image.

    Character profile of Mom, a woman in her late forties in a dark plum blouse leaning on a kitchen counter, from Xenogenesis
    Gertywood, director of Xenogenesis · GPT Image 2

  2. Step 2

    Make the line

    Record the line yourself, or write it in Text to Voice with Eleven v3 on the Free plan, and keep the take whose pace fits the character, since the avatar follows the audio's timing.

    00:04
    Gertywood, director of Xenogenesis

  3. Step 3

    Run Character Dialogue

    Attach the picture as Character Image and the line as Audio, pick Kling AI Avatar v2 or Kling AI Avatar v2 Pro, and use the Prompt for the expression, as Ulysses did for this short, strained line.

    Extreme close-up of a red-faced man with stubble straining to speak against a pale blue background, from Ulysses00:02
    Jaime Yao, director of Ulysses · Kling AI Avatar v2 Pro

  4. Step 4

    Reuse the voice

    Attach the same clip to every later Omni Reference shot and name it as the character's voice, as Xenogenesis did with Mom Voice and Cole's Voice at the family dinner.

    A pale young man in a grey sweatshirt sits between his parents at a dark dinner table, from Xenogenesis00:13
    Gertywood, director of Xenogenesis · Seedance 2.0 · Reference to video
  • Character Dialogue runs on the Standard plan ($20 a month) and is billed by the second of audio.

How do you make a picture talk from a typed line?

Use the picture as the start frame in Image to Video, quote the line in the prompt and keep audio on, so the model makes the voice and the lip movement together.
Extreme close-up of a red-faced young guard in a chainmail hood against a sky-blue background
The same guard's face speaking two lines straight into the camera00:05
The still from Ulysses, then the Kling O3 Pro shot that speaks two lines from itJaime Yao, director of Ulysses · Nano Banana 2 · Kling O3 Pro · Image to video

The shot's prompt

handheld camera style. Extreme close-up of his face. His face is flushed red and tense, and his eyes stay locked directly on the camera the whole time, with no eye drift and no looking away. He first says, “Still, here.” Then he pauses briefly, still staring into the camera, and adds, “I thought when you were hanged, you couldn’t talk.”
Jaime Yao, director of Ulysses

One portrait, many lines

One portrait can speak line after line, because each new prompt keeps the face and changes the words in quotes, as The Miracle did with its announcer.
The announcer's identity portrait: a silver-haired man in a grey uniform on a CRT television screen
The same announcer on a CRT broadcast, pausing before he says Correction. Three remain.00:15
The Miracle's identity portrait for its announcer, then the Seedance 2.0 shot where he says “Correction. Three remain.”Sam Wasserman, director of The Miracle · Seedance 2.0 · Image to video
12 clips The Miracle made from that one identity portrait, by the line each prompt asks the announcer to say word for word
The line, from the promptTakesMade with
Good evening. You are watching the final running of the Gauntlet. The oldest and most merciless trial still alive.1Seedance 2.0
The rules have never changed. Reach the tower. Reach the room with the glowing light. And stay alive long enough to reach the destination without dying.3Seedance 2.0, MiniMax H3
We have entered the final stretch of the final Gauntlet. Seven Runners remain. Four have reached the tower.3Seedance 2.0, MiniMax H3
The Sentinel.1Seedance 2.0
I do not believe what we have just witnessed. The Sentinel is down. For the first time… the way is open.1Seedance 2.0
He’s in.1Seedance 2.0
Correction. Three remain.1Seedance 2.0
Ladies and gentlemen… history has a familiar face. The last living Runner has reached the summit. His father stood in that same rain. He did not survive what came next.1Seedance 2.0
  • Each of the 12 prompts describes the same low broadcast voice with "faint old Mid-Atlantic precision" and asks him to "Speak exactly these words, in this order, with exact lip sync", so the announcer sounds like one man.

How do you keep a talking avatar's voice the same?

Attach the character's voice clip as an audio reference in Omni Reference and name it beside the line, so every new shot speaks in that one voice.
Handheld close-up of a pale young man with messy dark hair answering someone off screen at night00:06
Gertywood, director of Xenogenesis · Seedance 2.0 · Reference to video
Use Cole Character S1 Rehash as the visual reference for Cole.
Use Cole's Voice as what Cole's voice sounds like.
use Hf 20260722 031459 F6c056b3 8c51 4073 8651 E57006a8b386as the door of the house.
Identity lock — Cole has heterochromia: his right eye (on the viewer's left) is blue-grey, his left eye (on the viewer's right) is brown. Do not normalize both eyes to the same color.
Single continuous shot.
handheld shaky close up on Cole as he says with annoyance, "What?! Please no. Mom, I'm 18."
Blocking: Cole tight in frame standing 10 feet away from the door of the house, eyeline directed off-screen slightly off center left toward his parents.
Performance: Cole's irritation comes from being coddled too much by his mother.
Camera: handheld shaky tight close-up on Cole, holding as he listens and reacts.
Style: handheld, shaky camera, cinéma vérité. 35mm film emulation, halation, bloom, shallow depth of field with hexagonal bokeh.
Audio: no music, no score. Quiet night street ambience, Mom's and Dad's off-screen lines, Cole's dry irritated reply.
Forbidden: no white blobs on the character's faces.
  • In this shot from Xenogenesis, Cole's face comes from one image and his voice from one clip, Cole's Voice, which Gertywood cites in 142 of the film's 420 clips.
  • For making the voice itself and keeping it across a film, see the AI character voice generator guide.

What makes a talking avatar look real?

The prompts behind these clips fix the exact words, describe how each line sounds and keep the soundtrack clear, so the model spends its effort on the mouth.

  1. 1. Lock the words

    Tell the model to speak the quoted words exactly, because a video model left to itself can add, skip or reword a line and the lips then match the wrong words.

    Do not add, omit, repeat, paraphrase, or caption any word.
    Sam Wasserman, director of The Miracle

  2. 2. Hold the camera still

    Rule out camera moves when the face is the whole shot, because a still frame keeps the eyes and mouth sharp and makes any sync error easy to see.

    No camera movement or edit.
    Sam Wasserman, director of The Miracle

  3. 3. Say how the line sounds

    Give each line a tone, a pace and a volume, the way a director would, since a flat request gets a flat read from the model.

    In a soft, gentle, reassuring voice, he says, “Would you like to tell me his whole story?”
    Jaime Yao, director of A Proper Introduction

  4. 4. Time the line

    Give the second the line starts and the pauses inside it, so a short line lands where the cut needs it instead of filling the clip with talk.

    At approximately 00:07.5 say “Correction.” Let a sober one-second pause land. Then say “Three remain.”
    Sam Wasserman, director of The Miracle

  5. 5. Fix where the eyes go

    Decide whether the avatar looks into the lens or at someone off screen, because eyes that wander read as a glitch more than a stiff mouth does.

    his eyes stay locked directly on the camera the whole time, with no eye drift and no looking away.
    Jaime Yao, director of Ulysses

  6. 6. Keep the soundtrack clear

    Rule out music and talk over talk, so the voice stays clean enough to judge the sync and to cut against your own score later.

    Audio: no music, no score.
    Gertywood, director of Xenogenesis

What do people ask about AI talking avatars?

What is an AI talking avatar?

An AI talking avatar is a video made from a still picture of a person or character, where the face speaks a line with lip movement that matches the words. In Flick you make one with Character Dialogue from a picture and an audio line, or with Image to Video from a picture and a typed line.

Can I make a photo talk with AI for free?

Yes, you can start free: Text to Voice with Eleven v3 and Image to Video with MiniMax H3 Max, which makes video with synchronized audio, are on the Free plan. You get 100 credits when you sign up and 100 credits every week, while Character Dialogue needs the Standard plan at $20 a month.

How do I make an AI avatar talk in my own voice?

Record the line yourself, upload it as an mp3, wav, m4a, ogg or aac file, and attach it as the Audio in Character Dialogue next to your picture. The avatar then speaks with your exact timing and voice.

How does AI lip sync work for a talking avatar?

Character Dialogue moves the mouth and face to the sound of the audio you give it, so the timing comes from your recording. Image to Video models such as Seedance 2.0 make the voice and the mouth together from the words quoted in the prompt.

Should I use Character Dialogue or Image to Video for a talking avatar?

Use Character Dialogue when the line is already recorded and must stay exact, and Image to Video when you want the camera, the body and the voice made with the shot. Ulysses, a short made in Flick by Jaime Yao, uses both.

Can I make a photo of a real person talk?

Only with that person's permission, and never to put words in their mouth that they did not say or would not agree to.

Who owns the talking avatar videos I make?

You do. Content you generate on Flick is yours for personal and commercial projects. See the terms of use for detail.

More on making characters speak in Flick

See all

Films on this page

Give your character a voice and a face

One picture, one line, and the character speaks it. Keep every take, voice and shot on one canvas.

Make a talking avatar in Flick

Try for free, no credit card required.

Ask AI about this page:ChatGPTPerplexityClaude