DeepSeek Launches AI Imaging Toolkit

DeepSeek, the hottest name in the AI ​​market today, has just released multimodal AI models, claiming to beat OpenAI's Dall-E 3 image generator.

The models, available for download from the Hugging Face platform, are part of a new set of models that DeepSeek calls Janus-Pro. They range in size from 1 billion to 7 billion parameters. The larger the number of parameters, the better the model performs.

Comparison of Janus-Pro and Janus's ability to generate images from text. Photo: DeepSeek

Janus-Pro can analyze and generate new images. According to DeepSeek, on two AI benchmarks GenEval and DPG-Bench, Janus-Pro-7B beats Dall-E 3 as well as other models such as PixArt-alpha, Emu3-Gen, and Stable Diffusion XL.

However, according to TechCrunch, most of the remaining Janus-Pro models can only analyze small images, with a maximum resolution of 384 x 384. Still, the Janus-Pro's performance is impressive, considering its compact size.

DeepSeek, a Chinese startup founded in 2023, has suddenly attracted attention in recent days after its chatbot rose to the top of the App Store rankings in the US. The startup's large language models, trained using computationally efficient and cost-effective techniques, have prompted Wall Street to question whether the US can maintain its lead in the AI ​​race and whether demand for AI chips is sustainable.

On January 27, DeepSeek said it would temporarily restrict user registrations due to “large-scale malicious attacks” on its services. Existing users will still be able to log in as usual.

(According to TechCrunch)

Source: https://vietnamnet.vn/deepseek-ra-mat-bo-cong-cu-tao-anh-ai-2367126.html