Trending

Janus

deepseek-aiMIT

Multimodal17.7K Stars2.2K Forks12 views

DeepSeek's unified multimodal understanding and generation model series. Features decoupled visual encoding for both image comprehension and text-to-image generation within a single transformer architecture. Available in 1.3B, Pro-1B, and Pro-7B variants.

Key Features

Unified architecture for both vision-language understanding and image generation
Decoupled visual encoding optimizing each modality independently
Multiple model variants: Janus-1.3B, Pro-1B, Pro-7B
Classifier-free guidance for enhanced visual generation quality
JanusFlow variant with rectified flow for improved generation

Open Source

Janus

Key Features

Tags

Related Projects

AIRI