High-priority alert based on source significance and keyword relevance.
Context
What happened: Intern Lumina U2 is presented as a diffusion language model using multi-codebook discrete visual tokens to unify image, video, and 3D generation with text in one backbone.
Key numbers: 16.9B MoE diffusion LLM, 1.4B activated parameters, 8 AToken codes per visual position, 4.45x image generation speedup, trained on
Why it matters: The model claims leading performance among unified models in visual understanding, image generation, and image editing, beating the same-backbone LLaDA2.0-Uni on 8 of 10 benchmarks and surpassing Qwen-Image with about 14x fewer activated parameters.
What changed
Intern Lumina U2 is presented as a diffusion language model using multi-codebook discrete visual tokens to unify image, video, and 3D generation with text in one backbone.
Why it matters
The model claims leading performance among unified models in visual understanding, image generation, and image editing, beating the same-backbone LLaDA2.0-Uni on 8 of 10 benchmarks and surpassing Qwen-Image with about 14x fewer activated parameters.