---
language:
- en
- zh
library_name: transformers
pipeline_tag: video-text-to-text
tags:
- video-language-model
- vision-language-model
- multimodal
- video-understanding
- image-understanding
- streaming-video
license: apache-2.0
---

# VideoChat3-4B

VideoChat3-4B is the 4B-parameter release of VideoChat3, a fully open, efficient, and generalist Video Multimodal Large Language Model for video understanding. 
The model is trained with a scalable video data synthesis and curation pipeline covering general, long-form, and streaming video scenarios. VideoChat3 achieves strong performance across mainstream video understanding benchmarks and surpasses prior open-source models at comparable or larger parameter scales, while releasing model weights, training code, training strategy, and training datasets to support reproducible open research.

[📄 Paper](http://arxiv.org/abs/2607.14935) · [🌐 Homepage](https://mcg-nju.github.io/VideoChat3/) · [💻 GitHub](https://github.com/MCG-NJU/VideoChat3) · [🤗 Paper Page](https://huggingface.co/papers/2607.14935)

<p align="center">
  <img src="https://raw.githubusercontent.com/MCG-NJU/VideoChat3/main/docs/public/paper/overview.png" alt="VideoChat3 Overview" width="100%">
</p>

## Installation

```bash
pip install torch transformers accelerate qwen-vl-utils
pip install decord opencv-python-headless
```

For faster inference, FlashAttention 2 can be used if it is correctly installed in your environment:

```bash
pip install flash-attn --no-build-isolation
```

## Quick Start

The following examples show how to run image and video inputs. For a complete runnable script, please refer to `demo_vc3.py`.

### Image/Video Inference

```python
from transformers import AutoModelForCausalLM, AutoProcessor
from qwen_vl_utils import process_vision_info

model_id = "MCG-NJU/VideoChat3-4B"
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True,
)
model.eval()

processor = AutoProcessor.from_pretrained(
    model_id,
    trust_remote_code=True,
)

# image
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "/path/to/image.jpg"},
            {"type": "text", "text": "Describe this image."},
        ],
    }
]
# video
# messages = [
#     {
#         "role": "user",
#         "content": [
#             {"type": "video", "video": "/path/to/video.mp4"},
#             {"type": "text", "text": "What happens in this video?"},
#         ],
#     }
# ]

text = processor.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)

images, videos, video_kwargs = process_vision_info(
    messages,
    image_patch_size=14,
    return_video_kwargs=True,
    return_video_metadata=True,
)
inputs = processor(
    text=text,
    images=images,
    videos=videos,
    do_resize=False,
    return_tensors="pt",
    **(video_kwargs or {}),
)
inputs = inputs.to(model.device)
if hasattr(model, "dtype"):
    inputs = inputs.to(model.dtype)
generated_ids = model.generate(
    **inputs,
    max_new_tokens=128,
    do_sample=False,
)
generated_ids = [
    output_ids[len(input_ids):]
    for input_ids, output_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.tokenizer.batch_decode(
    generated_ids,
    skip_special_tokens=True,
    clean_up_tokenization_spaces=False,
)
print(output_text[0])
```


### Online Proactive Response 

VideoChat3-4B supports proactive response for streaming video understanding. 
The proactive response protocol uses three explicit states:

- `</Silence>`: the current frame is irrelevant to the user question, or no meaningful event has started yet.
- `</Standby>`: a relevant event is unfolding, but the model needs more future frames before giving a reliable answer. When `</Standby>` is predicted, the next frame is processed with a larger visual budget.
- `</Response>`: the model has collected enough visual evidence and provides the answer.

Example usage with the provided proactive demo:

```bash
python demo_vc3_proactive.py \
  --model MCG-NJU/VideoChat3-4B \
  --video /path/to/video.mp4 \
  --question "your question"
```

## Citation

```
@misc{videochat3,
      title={VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding}, 
      author={Xinhao Li and Yuhan Zhu and Xiangyu Zeng and Yuhao Dong and Haoning Wu and Zhiqiu Zhang and Yuandong Yang and Changlian Ma and Qingyu Zhang and Yansong Shi and Xinyu Chen and Haoran Chen and Zizheng Huang and Jun Zhang and Kun Ouyang and Lin Sui and Ziang Yan and Yicheng Xu and Chenting Wang and Yinan He and Hongjie Zhang and Yi Wang and Yu Qiao and Yali Wang and Ziwei Liu and Kai Chen and Limin Wang},
      year={2026},
      eprint={2607.14935},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2607.14935}, 
}
```