InfiniteTalk: Talking Head and Avatar Video Generation
InfiniteTalk by InfiniteYou is a talking head and avatar video generation model built on Wan2.1. Creates lip-synced talking avatars in 14B fp16 from a single image.
InfiniteTalk
Wan2.1Talking HeadAvatarLip SyncAudio-DrivenTalking head and avatar video generation model built on Wan2.1. Supports single and multi-character variants in 14B fp16. Creates natural lip-synced talking avatars from a single portrait image.
| Developer | InfiniteYou (ByteDance) |
| Architecture | Wan2.1-based I2V fine-tune |
| Model Sizes | 14B fp16 |
| Variants | Single character, Multi character |
| Capabilities | Audio-driven talking head, lip-synced avatar generation |
Overview
InfiniteTalk by InfiniteYou is a talking head and avatar video generation model fine-tuned on Wan2.1 I2V. It transforms a single portrait image into a natural, lip-synced talking avatar driven by audio input. The model supports both single-character and multi-character variants, enabling dialogue scenes with multiple speakers.
Built on Wan2.1 I2V 480p, InfiniteTalk preserves facial identity while generating synchronized mouth movements and natural head motions that match the input audio.
ComfyUI Integration
InfiniteTalk is compatible with ComfyUI WanVideo native nodes and Kijai's WanVideoWrapper. Model files are available on the Kijai/WanVideo_comfy repository in fp16, fp8, and GGUF formats.
Guides and workflows related to this model series.