Deepseek has already returned to another release, excluding the success of the new AI model virus on the weekend. This time, we are focusing on AI image generation.
China’s startups have released an image generation model called JANUS-PRO, aiming for a stable diffusion with US rival Dall-E 3. Multi -modal models that generate images from text prompts are said to exceed the services that compete in fields such as image quality and accuracy.
This release has created waves with a highly logical response of lightning, following the release of DeepSeek’s R1 model, and has been trained more quickly and more training. According to what is said, this model is running on a very advanced NVIDIA chip and questions how China is competing without accessing cutting -edge US technology. This app has recently surpassed Chatgpt by downloading the Apple App Store.
The back -to -two back release signals Chinese pushes to get a scaffold in a growing AI arm race. Meanwhile, last week, President Donald Trump pledged a up to $ 500 million partnership with Openai and other high -tech companies, and announced a new AI infrastructure.
This timing is also in line with the increase in scrutiny of high -tech companies in China and is more nervous than Tactoku’s data privacy concerns.
Read more: Deepseek is a Hot AI app. Don’t be too attached to it
JANUS-PRO is now downloaded to a face embracing the AI developer platform.
In the introduction of the download page, DeepSeek states: Next -generation unified multi -modal model. “
The size of this model is in the range of 1 billion to 7 billion parameters and is an important factor in the problem solving function.
The company calls JANUS-PRO a “new auto-framework”, separates procedures for analyzing and generating images, and using a single unified system to solve the previous task. I will do it.
“Deepseek not only reduces the conflict between the roles of the visual encoder in understanding and generating,” says Deepseek.
