---
title: Bernini-R-S2V Talking Head
emoji: 🎤
colorFrom: red
colorTo: purple
sdk: gradio
sdk_version: 6.20.0
app_file: app.py
short_description: Portrait + speech audio into a lip-synced talking video
python_version: "3.12"
startup_duration_timeout: 1h
models:
  - rzgar/Bernini-R-S2V
  - tolgacangoz/Wan2.2-S2V-14B-Diffusers
---

# 🎤 Bernini-R-S2V — Talking Head

Animate a **portrait image + speech audio** into a **lip-synced talking video**.

[Bernini-R-S2V](https://huggingface.co/rzgar/Bernini-R-S2V) is a speech-driven
video extension of Bernini-R, built on **Wan 2.2 S2V**. It adds single-speaker
audio conditioning so a still portrait can be driven to speak or sing in sync
with an audio track.

## How it works

- **Input:** a clear front-facing portrait + a mono speech clip (best results at
  ~3–10 seconds of clear speech).
- **Output:** an MP4 of the portrait animated to speak, with the audio muxed in.
- Runs on **ZeroGPU** with the community `WanSpeechToVideoPipeline`
  (diffusers PR #12258).

## Note on weights

The Bernini-R-S2V release ships ComfyUI single-file checkpoints (dual high/low
noise Wan 2.2 MoE) plus a wav2vec2 audio encoder, with no diffusers-format
export. The community diffusers Wan 2.2 S2V pipeline uses a single-transformer
S2V architecture, so to deliver the same self-hosted image + audio → talking
video experience this Space runs the **base Wan 2.2 S2V-14B** weights
(`tolgacangoz/Wan2.2-S2V-14B-Diffusers`) that Bernini-R-S2V is finetuned from.