Janus-Pro (1B, 7B)
Scaled-up Janus unified model (1B, 7B) with better training strategy and data, improving text-to-image instruction following and multimodal understanding.
Same decoupled-encoder design as Janus with a three-stage training strategy fix, more data (including synthetic aesthetic data) and larger sizes; released the day DeepSeek's app dominated headlines.
- Date
- Monday, 27 January 2025
- Lab
- DeepSeek
- Kind
- open-weights
- Access
- open weights (restricted license)
GitHub lists 2025-01-27; arXiv v1 2025-01-29.
Sources
This record was checked against its sources on 6 October 2026. How we check