Today's featured cutting-edge AI information, welcome to read 👇
💼 Alibaba launches B2B AI search application Accio, supporting 5 languages, helping buyers quickly match suitable vendors and products through multi-dimensional information analysis.
🎨 DeepSeek open-sources JanusFlow multimodal model, integrating image understanding and generation capabilities, surpassing mainstream models like Stable Diffusion and SDXL in multiple benchmark tests.
👗 Researchers release Fashion-VDM virtual try-on video generation model, requiring only clothing images and human videos to generate natural and realistic try-on effects while maintaining original actions and expressions.
Latest News
1. Alibaba launches AI search application: Accio, similar to Perplexity.
Primarily targeting B2B, it precisely matches suppliers, customization scope, pricing, retail sales volume, customer reviews, and other multi-faceted information. Buyers only need to input their requirements to get matching vendors or products.
Official website: https://www.accio.com/
Currently only supports 5 languages: English, German, French, Spanish, and Portuguese. Interested users can try it out.
2. DeepSeek AI open-sources unified multimodal model: JanusFlow.
Capable of handling both image understanding and generation tasks, including processing complex backgrounds and human figures in different poses. The generated images show impressive quality and semantic consistency.
GitHub: https://github.com/deepseek-ai/Janus
Furthermore, it has surpassed open-source image models like Stable Diffusion v1.5, SDXL, and LLaVA-v1.5 in multiple benchmark tests.
Cutting-edge Technology
1. A video generation model for virtual try-on: Fashion-VDM.
Only requiring a try-on clothing image and a human video, it can make the person in the video wear the try-on clothing, with rich and natural-looking clothing details.
Detailed introduction: https://johannakarras.github.io/Fashion-VDM/
Additionally, the video maintains consistent human actions, expressions, and body postures. The code is not yet open-sourced, but from the demonstration effects shown, it looks very impressive.
