DeepSeek recently launched its DeepSeek R1 AI model that is said to match the performance of OpenAI’s o1. Now, the Chinese AI startup announces a new AI model for image generation: Janus-Pro-7B. According to DeepSeek, this model surpasses the performance of similar models such as OpenAI’s DALL-E 3 and Stability AI’s Stable Diffusion. The model is available in two variants: with 1.5 and 7 billion parameters.For a demo version of the open-source model, users can visit HuggingFace.
Janus Pro 7B
The Janus-Pro-7B model works in a similar way to other image generation models. Based on text input, the model produces an image. According to DeepSeek, the model also offers image analysis generation capabilities. With it, users can have a caption added to the image or ask the model what it sees in the image.
DeepSeek used a ‘novel autoregressive framework that unifies multimodal understanding and generation’, according to the company. DeepSeek further claims that the Janus model outperforms existing models, including OpenAI’s DALL-E 3 and Stability AI’s Stable Diffusion. Janus-Pro-7B is a free open-source model. You can find a demo of the AI model on HuggingFace.