DeepSeek, a Chinese AI startup, has announced that its Janus-Pro-7B model has outperformed industry leaders such as OpenAI‘s DALL-E 3 and Stability AI’s Stable Diffusion in image generation benchmarks.
The new model, an upgrade from DeepSeek’s earlier Janus version, boasts improved image stability and richer details, setting a new standard for text-to-image generation.
Key Features of Janus-Pro-7B
Training Enhancements: Janus-Pro-7B benefits from refined training processes, better data quality, and a larger model size.
Synthetic and Real-World Data: The model adds 72 million high-quality synthetic images, balanced with real-world data, to produce visually appealing outputs.
Larger Model: With up to 7 billion parameters, Janus-Pro-7B achieves faster training speeds and improved task comprehension.
This breakthrough follows the success of DeepSeek’s DeepSeek-V3 assistant, which became the top-rated free app on Apple’s App Store in the U.S.
Leave a comment