Skip to main content
Find the section that matches what you’re seeing. If nothing here fixes it, contact support.
On current plans, this feature uses AI credits. Learn more about tracking your media minutes and AI credits.Legacy and Sunset plans track usage differently. See Understanding your Legacy and Sunset plan for details.

Generation and timing

Generation starts only after you click Generate avatar, and it runs in the background. Longer videos take longer to process. You don’t need to stay in the project; closing the tab won’t cancel it. Descript emails you when your avatar is ready to review.
If you just assigned an avatar or made a script edit, click Generate avatar in the Scene Editor to start generation.
Generate avatar button in the Scene Editor
Longer compositions can take a while, and you’ll get an email when it’s done. If it seems stuck, click Generate avatar again from the Scene Editor. If it still won’t complete, contact support.

Regenerating after edits

Any change to your script, speaker, or assigned avatar means the avatar has to be regenerated. This includes:
  • Editing the script or transcript
  • Switching AI speakers
  • Changing the underlying audio recording
  • Replacing or deleting the avatar image
Click Generate avatar again to re-sync the animation.
Each time you generate or regenerate, Descript debits AI credits based on your total length of spoken audio, not just the parts where the avatar is visible. To avoid spending credits twice, finalize your script and speaker labels before you generate. Check your balance in the Usage tab of your account settings.

Photo upload errors

This appears when Descript’s automated safety system flags your uploaded photo, most often for:
  • Celebrity or public-figure likeness
  • Inappropriate content such as nudity
Upload a different image with a human subject, facing forward, with neutral lighting and no alterations. See avatar best practices for full image requirements.
Avatar uploads support JPEG, PNG, and WEBP only. If your file is another type (like GIF or HEIC), convert it to a supported format and try again.
Avatar uploads support image files up to 10 MB. Reduce your file size and upload again.

Content and generation errors

This means your script was flagged during generation, a separate check from the image safety check that runs at upload. Moderation is handled by our model provider, and Descript doesn’t control the specific criteria. Revise your script, remove any graphic or offensive content, and generate again. If it persists, contact support.

Avatar visibility

Avatar visibility depends on layer order and visibility. Confirm the avatar layer is turned on and sitting above other media in the Properties panel. See layer visibility and order.
Timeline exports (XML, AAF) to tools like Premiere Pro or Final Cut don’t include avatar layers. This is expected: avatars must be baked into a video file first. Export your composition as a video (like MP4) from Descript, then import that into your external editor. See timeline exports.

Quality and known limitations

Avatar models can produce visual anomalies. Some are expected; here’s what you might see and what helps.
You might notice flickering lines or shapes near the bottom of the frame or around an avatar’s hands. To reduce them:
  • Add a gradient or solid-color background
  • Add a captions layer
  • Crop the frame slightly to remove the affected area
Avatar models may occasionally produce:
  • Background movement: complex backgrounds can animate along with the avatar. Use a simple solid-color background to avoid it.
  • Hand artifacts: hands may sit in awkward positions or move unnaturally.
  • Facial behavior during silence: subtle head swaying or lip movement when the avatar isn’t speaking.
  • Attitude prompts: a prompt applies to each segment of your composition. For example, “starts skeptical, ends smiling” makes the avatar skeptical at the start of each paragraph and smiling at the end.
If artifacts persist, try a different avatar, switch models in the model picker, or minimize long pauses between speech.
Avatars made with the Kling models may include hand movement as part of generation. See avatar models for what’s available.