Changelog
Witness Janus-AI's Innovation Journey in Multimodal AI
11 months ago
- 🎉 First Release: Janus - The first multimodal model based on DeepSeek-LLM-1.3b-base and SigLIP-L
- ✨ Features: Visual encoding for multimodal understanding and generation
- 📜 MIT License for full commercial use
11 months ago
- 🚀 Released JanusFlow - A unified multimodal understanding and generation model
- 📊 Surpassed LLaVA-v1.5 and Qwen-VL-Chat in benchmark tests
- 🖼️ Support for 384×384 image generation
10 months ago
- 🌐 Launched online demo for Janus
- 🔄 Added support for DeepSeek Janus and Qwen2-VL
- 💻 Browser-based pure inference experience
8 months ago
- 🎯 Released Janus-Pro 1.3B & 7B models
- 🔥 Surpassing previous unified models in performance
- 🛠️ Enhanced framework flexibility and efficiency
- 🌟 Advanced visual encoding decoupling architecture