DeepSeek has Released Janus-Pro-7B, Surpassing DALL-E 3 and Stable Diffusion in Image Generation Capabilities

Chinese company DeepSeek has introduced its new multimodal neural network, Janus-Pro-7B, designed for efficient image recognition and generation. Compared to previous versions, the new model has been significantly improved: training algorithms have been optimized, the dataset expanded, and the model size increased. These changes have allowed the neural network to greatly enhance its performance in tasks related to image generation from textual descriptions and image analysis.

Janus-Pro-7B has outperformed well-known models like DALL-E 3 and Stable Diffusion in GenEval and DPG-Bench tests. This neural network offers high levels of detail and realism in image creation, making it a strong competitor against other solutions. The model is already available on the Hugging Face platform, and the code can be downloaded for free from GitHub, enabling researchers and developers to contribute to its development.