From the source
From the source
Unified continuous tokenizer for vision understanding and generation
Ant Ling introduced Ming-UniVision, an autoregressive vision-language model built on the MingTok tokenizer, which is the first visual tokenizer based on a continuous latent space.
The unified representation supports image understanding, generation, and editing within a single next-token prediction framework, achieving over 3.5× faster training convergence compared to traditional approaches.
From the source
MingTok is the first visual tokenizer based on a continuous latent space. It delivers a truly unified and efficient representation that serves as the bedrock for Ming‑UniVision’s unified NTP (Next‑Token Prediction) framework—harmonizing image understanding, generation, and editing in one in‑context multimodal loop.
developer.ant-ling.com
