Deepseek, the Startup China who has triumph in the US and beyond, presents Janus-Pro 7b. According to reports from Techcrunchin the last hours the set of multimodal AI models surpassed the Dalle-3 image generator of the OpenAi rival. It is now available for download from the Hugging Face development platform; is covered by a MIT license from software Free, which means that it can be used without restrictions.
X content
This content can be viewed on the site it Originates From.
What is Janus-Pro
The company of AI calls Janus-Pro 7B a “new self-representive framework”, capable of understanding and creating images. Unlike other “unified models”, visual coding decouples for interpretation and multimodal generation, using the Siglip-L model as a encoder and a Tokenizeran element used to translate text into data that can be processed by the model, of the llamagen set. This is an important innovation that allows you to overcome the performance of some of the most popular models in the market such as Dalle-3, Pixart-Alpha, Emu3-Gen and Stable Diffusion XL.
With the MIT license, the user can use and modify the code freely, even for commercial purposes, as long as the original copyright notice is maintained. Within AI models, it is one of the most permissive concessions that exist. However, Janus-Pro 7B requires accepting the Deepseek license, which includes ethical restrictions such as the prohibition of military use or the generation of content inclined to misinformation.
How does it work?
However, not everything that shines is gold: the new set of Deepseek models also has its defects, starting with the fact that You can only analyze small images, with a maximum resolution of 384 x 384 pixels. Taking into account the small size of Janus-Pro 7B, its performance is worth mentioning, as demonstrated by the results of some evidence shared by the company in Hugging Face.
Its operation does not differ much from those currently available in the market: All you have to do is describe a photo or work of art, Janus-Pro 7B will be responsible for converting that description into reality. Deepseek has demonstrated once again that it is able to improve existing technology, making it more attractive and functional for its users, as it has done with its Chatbot of which it is already causing shock in US application stores. It is a detail that worries Silicon Valley and Startups of the sector who fear being overshadowed by the Chinese competitor.
Article originally published in Wired Italy. Adapted by Alondra Flores.
#Deepseek #Lanza #JanusPro #generator #images #competes #Dalle3