Skip to content

Virus AI Company Deepseek released a new image model series

    Virus AI Company Deepseek released a new image model series

    Virus AI Company Deepseek released a new image model series

    Virus AI company Deepseek released a new set of multi-mode AI models, claiming that it can surpass the Dall-E 3 of OpenAI.

    These models can be downloaded from the AI ​​Dev Platform hugging surface, which is part of the new model series of Janus-PRO in Deepseek. Their scale ranges from 1 billion to 7 billion parameters. The parameters are roughly corresponding to the skills of the model solving problems, and more models with more parameters are usually better than models with fewer parameters.

    Janus-PRO meets MIT permission, which means that it can be used in business without limit.

    DeepSeek image
    DEEPSEEK's image output of the Janus Pro model.Image source:Deepseek

    Deepseek describes it as the “new self-regression framework” Janus-PRO can analyze and create new images. According to the company, the largest Janus-Pro model Janus-PRO-7B In the two AI assessment benchons, Geneval and DPG Bench defeated Dall-E 3, and Pixart-Alpha, EMU3-Gen, and other model stability AI AI. Stable diffusion XL.

    Of course, some of these models are on the older side, and most Janus-PRO models can only analyze small images with a resolution of up to 384 x 384. But considering the compact size of these models, Janus-PRO's performance is impressive.

    DeepSeek wrote in an article in “Hugging Face”: “Janus-PRO exceeds the previous unified model and exceeds the performance of the mission model.” Make it a powerful candidate for the next generation of unified multi -model models.

    DeepSeek image
    Deepseek's new Janus Pro model is compared with the game.Image source:Deepseek

    DeepSeek is a Chinese artificial intelligence laboratory. It is mainly funded by quantitative trading company High-Flyer Capital Management. After its chat robot application rose to the top of the Apple App Store Chort, this week broke into mainstream consciousness. DeepSeek's language model uses computing efficient technical training, which allows many Wall Street analysts and technicians to question whether the United States can maintain its leading position in the AI ​​competition and whether the demand for AI chips can be maintained.

    Update: The early version of this story implies that the Janus-PRO model can only output small (384 x 384) images. That's not real. We regret wrong.


    TechCrunch has a news communication focusing on AI! Register here Put it in the inbox every Wednesday.


    Esportes duas vezes campeões todos os treinadores de campeões esportivos Caitlin Clark, Angel Reese Barron Trump: Do berço (literalmente) do ouro à vida do Senhor – último A bridge too far: Brisbane with millions of dollars in revitalization costs caught icon | Brisbane O presidente espanhol RTVE exige “paz e justiça da Palestina” ACC revenue and expenditure climbed in 2023-24. In FSU, Clemson lawsuits, so is legal bills Em 82 minutos (cronicos), há dois tabus do tempo e setenta anos caíram Chris Richards of the United States: FA Cup championship means the world to me “Eu acho que é o menor pecado para esta situação”: Rui Costa confia em Bruno Lage, embora incapaz de ganhar o campeonato – vídeo Warren Buffett