As if the launch of a new AI model that shaken the whole industry was not enough, the Chinese startup Deepseek followed this week by publishing a Image generator Ai He claims Provides “significant progress in multimodal understanding and monitoring capacity for text instructions in the image”.
The new image generation model is called Janus-ProAnd it aims to compete with us rivals like Dall-E 3 and stable diffusion. The new model claims to surpass its competition in fields such as image quality and precision.
The launch of Janus-Pro has only intervened a few days after the release of the DEEPSEEK R1 model, which made waves with its very logical and very logical logical responses, and for being formed more quickly and a fraction of the cost of the models Americans.
The Deepseek model would have taken place on less advanced Nvidia chips, which raises questions about how China is in competition without access to advanced American technology. The iOS application has exceeded the ChatPPT in downloads on the Apple App Store, and is still the free application n ° 1 to January 31.
The consecutive versions point out the thrust of China to gain a place in the arms race of the growing AI. Meanwhile, last week, President Donald Trump announced a new AI infrastructure initiative, committing to $ 500 million in partnership with Openai and other technological companies.
Look at this: What is Deepseek Ai? Everything that needs to know about the new popular AI
The publication of R1 and Janus-Pro also coincides with an increased examination of Chinese technological companies, with already high tensions on Tiktok data confidentiality problems.
In an introduction to his download page, Deepseek declares: “Janus-Pro goes beyond his previous unified model and corresponds or exceeds the performance of specific models. The simplicity, high flexibility and the efficiency of Janus-Pro A solid candidate for new generation unified multimodal models.
The model varies in size from 1 billion to 7 billion parameters, a key factor in its problem solving capacities.
The company calls Janus-Pro a “new self-regressive framework” which resolves the previous challenges by separating the steps of analysis and generation of images, while using a single unified system to treat everything.
“The decoupling attenuates not only the conflict between the roles of visual encoder in understanding and generation, but also improves the flexibility of the frame,” wrote Deepseek.
The user’s response to Janus-Pro has been mixed so far, with Some redditors claim The images have been resembling the efforts of its competitors since the past years. To get an idea of how Janus-Pro is compared to the other IA images generators, see this performance ventilation between Chatgpt 4o, Qwen 2.5 and Janus-Pro de Youtuber Ejack Yao.
Janus-Pro is currently Available for download On the developer platform AI hugging the face.
