Alibaba's Mach Creates Lifelike AI Avatars from Text

Alibaba has developed a new text-to-3D model called Make-A-Character (Mach) that can generate detailed and lifelike 3D avatars from simple text descriptions.

Summary

  • Alibaba's Mach seamlessly converts textual descriptions into visual avatars, providing an easy way to create custom 3D avatars.
  • It leverages large language and vision foundation models like Stable Diffusion and ControlNet to generate reference portrait images from text.
  • The generated 2D images are then converted into 3D face meshes and textures and assembled with matched accessories.
  • The avatars focus on Asian ethnicities currently, but support for different ethnicities and styles will be added.
  • Expression, motion and cloth generation capabilities driven by text will also be developed.
  • Alibaba also introduced Richdreamer for 2D to 3D generation and Animate Anyone for transforming images into character videos.
  • Additionally, new language models Qwen-72B and Qwen-1.8B were launched to advance language capabilities.

READ MORE

Related post

Embeddings

Do Text Embeddings Truly Understand Meaning or Just Tokenize?

The key takeaway is that embeddings are vector representations that capture the semantics and meaning of the text, going beyond just tokenization. The embedding process squeezes the text through the model to understand it and make predictions, thus encoding semantic relationships. Better data quality can improve embeddings but normalization is…

ChatGPT

What to Expect from GPT-5

GPT-5 is OpenAI's upcoming AI chatbot that is expected to be more advanced than the current GPT-4 model that powers ChatGPT. It will likely have broader knowledge, better personalization, and multi-modal capabilities to process images, audio, etc. READ MORE