GitLab Transcend - Intelligent orchestration
Mini toolsAVTR 1
By GOODSIZE
AVTR-1 is a flow-matching-based autoregressive model from Avaturn Live for live dialogue avatars. Given a portrait image and dual-stream audio, it renders a talking avatar with lip-synced speech and active listening behavior at 25 fps on a single GPU. The repository includes model weights, inference code, an interactive streaming demo, and offline generation support for single-speaker speech, two-speaker dialogue, and idle motion. Its licensing is mixed: AVTR-1 model weights and scripts use the AVTR-1 Community License, while renderer and streamer components are noncommercial PolyForm-licensed unless separately licensed for commercial use.
New
Multimodal
Gen 3
Overview
AVTR-1 is Avaturn Live’s avatar video model for real-time lip-synced speech and active-listening animation from a portrait image plus audio streams.
About GOODSIZE
Create realistic 3D avatar with a selfie, customize, export as 3D model. Developer? Integrate our avatar SDK into your app or metaverse.
Industry:
Software Development
Company Size:
11-50
Location:
Wilmington, Delawere, US
Website:
avaturn.me
Tools using AVTR 1
No tools found for this model yet.
Last updated: May 26, 2026
