---
title: Bernini Edit · Style Transfer & Local Edits
emoji: 🎨
colorFrom: purple
colorTo: green
sdk: gradio
sdk_version: 6.15.0
python_version: "3.10"
app_file: app.py
pinned: false
license: apache-2.0
short_description: Restyle or locally edit a video with one instruction
models:
  - ByteDance/Bernini-R-1.3B-Diffusers
tags:
  - video-editing
  - style-transfer
  - wan2.1
  - zerogpu
  - ahead-of-time
---

# 🎨 Bernini Edit · Style transfer & local edits

Edit a **video** with a plain-language instruction using
[Bernini-R 1.3B](https://huggingface.co/ByteDance/Bernini-R-1.3B-Diffusers)
(ByteDance's Wan2.1-based renderer). Restyle the whole clip (*"make it a Van Gogh
painting"*, *"turn it into anime"*) or make a local edit (*"add a snowman on the
right"*, *"remove the watermark"*). Output keeps the source resolution and length.

The diffusion transformer blocks are served from a **precompiled ahead-of-time
(AOTI)** build via `spaces.aoti_load`, so there is no `torch.compile` on startup and
each generation is faster on ZeroGPU.

## Tips
- **Style transfer:** describe the target style for the whole frame.
- **Local edits:** name the object and the change, and state what to keep unchanged.
- **ω_V** controls how closely the source is preserved; **ω_TI** weights the instruction.