DeepSeek Introduces Janus Pro, a New AI Model for Image Generation

Chinese company DeepSeek presented a new model for image generation - Janus Pro. It is available in two versions with different parameters and, according to the company, can compete with other industry leaders. This was reported by TechCrunch.

Janus Pro, which DeepSeek describes as a "new autograde model," can both analyze and create new images. According to the company, on two AI evaluation platforms, GenEval and DPG-Bench, Janus Pro's largest model, Janus-Pro-7B, outperforms DALL-E 3, as well as models such as PixArt-alpha, Emu3-Gen, and Stable Diffusion XL from Stability AI.

Of course, some of these models are quite old, and most Janus Pro models can only analyze small images. However, the performance of Janus Pro is impressive considering the compact size of the models (from 1 billion to 7 billion parameters). Janus Pro is also licensed by MIT. This means that they can be used commercially without restrictions.

"Janus-Pro outperforms previous federated models and matches or exceeds the performance of task-specific models," DeepSeek writes in a post on Hugging Face. "Janus-Pro's simplicity, high flexibility, and efficiency make it a strong candidate for next-generation models that combine multiple modalities."

Test versions of Janus Pro are already available for download on the Hugging Face AI developer platform.

By the way, DeepSeek's success has attracted the attention of investors, as the shares of such tech giants as NVIDIA and Oracle have fallen. Even DeepSeek's main competitor, OpenAI, has recognized the impressive achievements of the Chinese startup.

You can read more about DeepSeek, its successes, and founders in our separate article.