How to Make a Talking AI Avatar You Can Reuse

Quick Answer: Design one avatar image, keep one voice for it, and reuse both. On Seedance, make the avatar talk with Lip Sync in Image mode for long presenter clips, or with Seedance 2.5 Reference to Video when you want the avatar speaking inside a new scene.

Most guides stop after the first talking clip. The harder part comes with episode two: the face drifts, the voice changes and the series stops looking like one character. This guide shows how to build a talking AI avatar once and keep it consistent across every video you make on Seedance.

If you want to animate a portrait into a short avatar clip without a voice, start with our guide to AI avatar videos with Seedance. This article is about avatars that speak.

The Three Parts of a Reusable Avatar

Save all three in one folder before you make the first video.

Step 1: Design the Master Avatar Image

Open Text to Image and describe the character in detail. Compare a few models, such as Nano Banana 2, Seedream 5.0 Pro and GPT Image 2.5, and pick one result you are happy to see in every episode.

Head-and-shoulders portrait of a cheerful man in his late twenties, short curly black hair, round glasses, mustard knit sweater, looking directly at the camera, soft daylight from the left, plain light-blue background, photorealistic.

Then check it:

• The face is sharp and faces the camera.
• The mouth and chin are fully visible.
• The background is plain, with no text or second person.

Using your own likeness instead? Start from a clear, recent photo of yourself. To try a new outfit first, edit it with Image to Image.

Step 2: Fix the Avatar's Voice

Create the voice in Text to Speech, which offers Gemini 3.1 Flash TTS and Seed Audio 1.0. Choose one voice and one delivery style and write them down, for example "Seed Audio 1.0, warm male voice, English, calm pace". Use the same settings for every script.

You can also record the scripts yourself or with a voice actor. Record in the same room with the same microphone each time so the character sounds identical from episode to episode.

Step 3: Choose How the Avatar Talks

Seedance gives you two ways to make the same avatar speak.

Option A: Talking photo for presenter clips

Open Lip Sync, switch to Image mode, and upload the master image and the voice track. This keeps the original framing and is best for longer explainers, lessons and updates.

Option B: Seedance 2.5 scene with your avatar

When you want the avatar walking through a store, sitting in a café or presenting at a whiteboard, use Reference to Video with Seedance 2.5. Add the master image as a reference, add the voice track as an audio reference, and describe the scene:

@1 the man in the mustard sweater stands behind a coffee counter, looks at the camera and says the line in the audio, then gestures to the menu board behind him. Medium shot, warm morning light, gentle push-in.

Seedance 2.5 accepts up to 30 reference images and up to 30 seconds of reference audio, generates up to 30 seconds per clip at 480p or 720p, and costs 8.5 credits per second. Keep scene lines short: one or two sentences per clip.

Step 4: Make Episode Two (and Ten)

1. Write the new script and create the audio with the saved voice settings.
2. Upload the same master image, not a regenerated one.
3. Reuse your style notes in the prompt and change only the scene and action.
4. Compare the new clip with an earlier one side by side before publishing.

If the face starts to drift in Option B, add a second reference image of the avatar, such as a three-quarter view made from the master portrait with Image to Image.

Fixing Common Problems

FAQ

What is a talking AI avatar?

A digital character that speaks your script on camera, created from one image and one voice track.

Which option should I start with?

Use Lip Sync in Image mode for presenter-style videos longer than 30 seconds, and Seedance 2.5 Reference to Video for short scenes where the avatar moves through a setting.

Can the avatar use my own voice?

Yes. Upload your own recording instead of a text-to-speech track.

How much does it cost?

A 30-second talking photo costs about 90 credits with Kling AI Avatar Standard or 150 with OmniHuman 2.0. A 10-second Seedance 2.5 scene costs about 85 credits. The tool shows the exact cost before you generate; see Pricing.

Can I make an avatar of someone else?

Only with their permission. Never use an avatar to impersonate a real person or mislead viewers.