Ming-Lite-Uni: Advancements in Unified Architecture for Natural Multimodal Interaction
Open-source framework for unified visual understanding and generation
Open-weight release
Ant Ling released Ming-Lite-Uni, an open-source multimodal framework that integrates a unified visual generator with a native multimodal autoregressive model for vision and language tasks.
The framework supports text-to-image generation and instruction-based image editing, achieving an average understanding score of 69.7 on OpenCompass and an image generation score of 0.62 on GenEval.
It is currently in alpha phase and available with full model weights and test code.
