SOTAMak1r/VINO-weight

VINO — Unified Visual Generator (Official Weights)

VINO: A Unified Visual Generator with Interleaved OmniModal Context

🌐 Project Page • 📑 Paper • 💻 Code • 📺 Demo Video

---

🔥 What is VINO?

VINO is a unified image & video generation and editing framework powered by a Vision-Language Model (VLM) and Multi-Modal Diffusion Transformer (MMDiT).

A single set of weights supports:

• Text-to-Image
• Text-to-Video
• Image-to-Video
• Multi-Image-to-Video
• Image Editing
• Video Editing
• Element Cloning

One model. All visual generation & editing tasks.

---

📦 Contents of This Repository

This Hugging Face repository provides the official VINO model weights, including:

• MMDiT backbone
• Learnable multimodal tokens

These weights are intended to be used with:

👉 https://github.com/SOTAMak1r/VINO-code

---

🧩 Required Base Models

VINO depends on the following public checkpoints:

Component	Source
VLM	Qwen/Qwen3-VL-4B-Instruct
Video VAE	hunyuanvideo-community/HunyuanVideo

They will be automatically downloaded by the VINO codebase.

---

⬇️ Download

Option 1: Hugging Face CLI

huggingface-cli download SOTAMak1r/VINO-weight \
  --local-dir ./checkpoints/SOTAMak1r/VINO-weight \
  --local-dir-use-symlinks False

Option 2: Inside VINO Repo (recommanded)

python download.py --ak YOUR_HF_TOKEN

---

🚀 Quick Start

See full instructions in:

👉 https://github.com/SOTAMak1r/VINO-code

---

📄 License

• Model Weights: CC BY-NC 4.0 (Non-Commercial Only)
• Code: Apache 2.0

---

📝 Citation

@article{chen2026vino,
  title={VINO: A Unified Visual Generator with Interleaved OmniModal Context},
  author={Chen, Junyi and He, Tong and Fu, Zhoujie and Wan, Pengfei and Gai, Kun and Ye, Weicai},
  journal={arXiv preprint arXiv:2601.02358},
  year={2026}
}