DeepSeek Releases Janus Pro a Powerful AI Image Generator
DeepSeek, a Chinese company best known for its AI chatbot, has announced a family of multimodal AI models called Janus Pro. The company says these algorithms, available under an open MIT license, can generate and analyze images better than OpenAI DALL-E 3, PixArt-alpha, and Stable Diffusion XL.
Launched on January 27, 2025, Janus-Pro-7B stands out for its impressive ability to process both text and images, giving users new tools for creative expression.
According to renowned AI analyst Rowan Cheung, the model actually outperforms its competitors in a number of ways. Janus-Pro-7B represents significant improvements over previous DeepSeek models. Using the innovative SigLIP-Large-Patch16-384 encoder, the neural network breaks down images for detailed analysis, preserving details for improved interpretation and synthesis.
As experts note, the revolutionary nature of Janus Pro is its versatility. It combines analysis and generation, which is rare in models of this size. Although the image resolution does not yet reach the level of DALL-E 3, the open license and adaptability can accelerate the implementation of Janus Pro in commercial projects.
