Avatars are animated presenters for your content. They can stand in for a talking head and speak using a stock AI voice or your custom voice clone. Just designate a speaker, assign an avatar, and Descript will generate an animated presenter to deliver your content onscreen.

Use a pre-made avatar from the Descript gallery or create your own by uploading a photo, generating an avatar image from a text prompt, or combining both. Learn how to create a custom avatar.

Avatars work with both text-to-speech (TTS) and your own recorded content. Each speaker in the composition can have one avatar assigned at a time, which applies consistently across the entire project.

This article covers

• Using avatars in your project
• Managing or updating an avatar
• Tips for using avatars
• Available avatar generation models

How to use avatars in your project

The workflow depends on the type of audio you’re using:

• Write a script and generate audio with text-to-speech
• Record your own voice directly in Descript
• Import pre-recorded audio, like a podcast or screen recording

For step-by-step instructions, see How to use avatars in Descript.

Manage avatars in a project

Once assigned, an avatar follows its speaker anywhere that speaker appears in your composition. You can:

• Update or remove avatars by clicking the speaker label and choosing Update speaker’s avatar
• Resize, reposition, and crop the avatar in the Scene panel
• Use avatars with visual effects like Green Screen
• Regenerate the avatar at any time (Note: changes to the script or speaker label require re-generation)

Tips for using avatars

• Generate after you’ve finalized your script and speaker labels
• Use Replace media to swap a camera layer for an avatar
• If an avatar isn’t visible, confirm it’s above other visuals in the layer stack; see layer order
• For Kling models: Use the attitude prompt to adjust your avatar's tone and expression. Shorter prompts tend to work best. Focusing on facial expression or posture can be helpful (e.g., 'warm smile, steady eye contact' or 'professional and calm'). To reduce hand gestures, consider avoiding energy words like 'enthusiastic,' 'animated,' or 'dynamic.'

Avatar generation models

Choose between three available avatar generation models. AI credit cost differs per model. Read more about AI credits.

Known limitations with AI avatar models

Avatar models may occasionally produce visual anomalies. Some things you might notice:

• Background movement: Complex backgrounds may be animated along with the avatar.  To avoid this, try using a simple solid-color background.
• Hand artifacts: Hands may appear in awkward positions, make mismatched gestures, or move unnaturally. 
• Facial behavior during silence: Avatars may show subtle head swaying, lip puckering, or other micro-movements when not speaking
• Attitude prompts: Prompts may be applied to each segment of your composition. For example, "starts out skeptical, but ends up smiling by the end" will make your avatar skeptical at the beginning of each paragraph and smiling at the end of each paragraph.

What you can try: While there’s no guaranteed fix, these adjustments may help reduce visible artifacts:

• Try a different avatar
• Test a different avatar model in the model picker
• Minimize long pauses between speech