Views
No views yet
| Field | Description |
|---|---|
| Model type | Full video generation/editing pipeline with an MLLM-based semantic planner and a DiT-based renderer. |
| Checkpoint | ByteDance/Bernini-Diffusers-v2 |
| Code | ByteDance/Bernini |
| Renderer base | Wan2.2-T2V-A14B |
| Planner base | Qwen2.5-VL-7B-Instruct |
| Recommended use | Complex generation/editing requests that benefit from explicit latent semantic planning and stronger instruction following. |
| Supported tasks | t2i, i2i, t2v, v2v, rv2v, and r2v. |
| Model behavior | Decomposes complex instructions and plans semantic changes before rendering, at the cost of a heavier checkpoint layout than Bernini-R. |
| Model | EditVerse | OpenVE | OpenS2V | VBench | Bernini-v2v (OS) | Bernini-rv2v (OS) |
|---|---|---|---|---|---|---|
| Bernini-v2 7+14B | 8.02 | 3.96 | 63.83 | 84.46 | 3.49 | 3.55 |
Bernini-Diffusers-v2 directory directly to --config.1Bernini-Diffusers-v2/
2 bernini/
3 mllm/
4 scheduler/
5 t5_text_encoder/
6 t5_tokenizer/
7 vae/
8 config.json
9 transformer_config.json
10 transformer_2_config.jsonbernini/ provides the Bernini planning checkpoint.mllm/ provides the Qwen2.5-VL planner assets.transformer_config.json and transformer_2_config.json define the Wan2.2 diffusion decoder components used by the full pipeline.t5_text_encoder/, t5_tokenizer/, vae/, and scheduler/ provide the base diffusion modules required for inference.1pip install -U "huggingface_hub"
2hf download ByteDance/Bernini-Diffusers-v2 \
3 --local-dir pretrained_models/Bernini-Diffusers-v21git clone https://github.com/bytedance/Bernini.git bernini && cd bernini
2pip install -r requirements.txt
3# Open-VeOmni is required. Install it with --no-deps so it does not pull in a
4# different torch build and override the pinned torch==2.7.1+cu126:
5pip install --no-deps git+https://github.com/ByteDance-Seed/VeOmni.git@v0.1.11--config:1python infer_single_gpu.py --config pretrained_models/Bernini-Diffusers-v2 \
2 --case assets/testcases/i2i/i2i.json --num_frames 1--use_pe enhances the prompt through an OpenAI-compatible endpoint and is recommended for best generation quality.1export BERNINI_PE_API_KEY=... # or OPENAI_API_KEY
2export BERNINI_PE_BASE_URL=... # or OPENAI_BASE_URL
3export BERNINI_PE_MODEL=... # vision-capable chat model1# Single GPU
2python gradio_demo.py --config pretrained_models/Bernini-Diffusers-v2 --port 7860
3
4# 8 GPUs, 8-way Ulysses sequence parallel
5torchrun --nproc-per-node 8 gradio_demo.py --ulysses 8 \
6 --config pretrained_models/Bernini-Diffusers-v2 \
7 --port 7860 --sharescripts/bernini_v2/ directory in the Bernini repo provides ready-to-run task launchers for Bernini-Diffusers-v2:run_t2i.shrun_i2i.shrun_t2v.shrun_v2v.shrun_rv2v.shrun_r2v.shrun_gradio.shexport BERNINI_CONFIG=/path/to/Bernini-Diffusers-v21@article{bernini,
2 title = {Bernini: Latent Semantic Planning for Video Diffusion},
3 author = {Chenchen Liu and Junyi Chen and Lei Li and Lu Chi and Mingzhen Sun and Zhuoying Li and Yi Fu and Ruoyu Guo and Yiheng Wu and Ge Bai and Zehuan Yuan},
4 journal = {arXiv preprint arXiv:2605.22344},
5 year = {2026}
6}