metadata
license: apache-2.0
base_model:
- ByteDance/Bernini-R
tags:
- text-to-video
- image-to-video
- video-to-video
- image-to-image
- audio-to-video
- wan-s2v
- bernini-r-s2v
Bernini-R-S2V
Speech-driven video on Bernini-R , T2V, I2V, and V2V with lip-sync.
This model adds single-speaker audio support to Bernini-R, so you can drive video with speech in text-to-video, image-to-video, and video-to-video setups. It is not state-of-the-art audio-to-video, but it removes the need for post-processing or extra models just to add speech to Wan videos. For basic talking-head work, or longer videos built from short sequences, it is a handy all-in-one option on top of Bernini's motion and editing strengths.
| File | Role |
|---|---|
| 'wan2.2_bernini_r_high_noise_fp16_s2v.safetensors' | High noise |
| 'wan2.2_bernini_r_low_noise_fp16_s2v.safetensors' | Low noise |
| 'wan2.2_bernini_r_high_noise_fp8_scaled_s2v.safetensors' | High noise |
| 'wan2.2_bernini_r_low_noise_fp8_scaled_s2v.safetensors' | Low noise |
| 'wan2.2_bernini_r_high_noise_int8_convrot_s2v.safetensors' | High noise |
| 'wan2.2_bernini_r_low_noise_int8_convrot_s2v.safetensors' | Low noise |
ComfyUI detects these as WAN22_S2V / 'WanModel_S2V' (audio keys trigger S2V model type).
Usage
- Download the checkpoints.
- Place in:
ComfyUI/models/diffusion_models/ - Add wav2vec2 to:
ComfyUI/models/audio_encoders/ - Install ComfyUI-WanBerniniS2V custom node:
(Create a folder in 'ComfyUI/custom_nodes/' named 'ComfyUI-WanBerniniS2V', then save the Python files in that folder.) - Restart ComfyUI.
- Search for the Bernini S2V Conditioning node or use the example workfllow
Audio tips
| Setting | Recommendation |
|---|---|
| Channels | Mono, wav2vec2 downmixes stereo internally |
| Sample rate | 44.1 khz or 48 khz (resampled to 16 kHz) |
| Content | Clear speech, less background music = better sync |
| Length | in my testing, max 9 to 15 seconds |