Model reference · open weights

avtr-1

avtr-1 is an open-weight video model from avaturn-live. avtr-1 (BF16) weighs 613 MB; the smallest configuration that runs it is RTX 3060 12 GB.

• AVTR-1 is an open-source real-time avatar model by avaturn-live that generates lip-synced speech and active listening video from a portrait image and dual-stream audio.
• It operates at 25 fps on a single GPU, with performance benchmarks ranging from 84 ms per 5-frame chunk on an L40 to 232 ms on an RTX 4060.
• The model weights are released under the AVTR-1 Community License, which permits commercial use for entities with annual revenue below USD 10,000,000, subject to specific use restrictions.

Summary of the avaturn-live/avtr-1 model card, 2026-10-01

Hugging Face ↗