AI Briefing#diffusion-model
The latest AI and developer news about #diffusion-model, with the original source and a short summary.
The latest AI and developer news about #diffusion-model, with the original source and a short summary.
NVIDIA has unveiled Nemotron-TwoTower-30B-A3B, a diffusion-based language model that boosts generation speed by 2.42x.
A framework has been released for unified evaluation of Diffusion Transformer model performance, spanning from ImageNet to T2I.
Netflix unveiled two research models designed to address the limitations of existing generative AI—unintended alterations and physical discontinuity.
Moebius is a lightweight inpainting framework that, through LλMI blocks and an adaptive distillation strategy, reduces parameter count to under 2% while delivering performance comparable to large-scale models.
Samsung R&D Institute has unveiled LAFUFU, a technology that dramatically speeds up high-resolution speech restoration by leveraging Latent Space.
A research study introduces an efficient Diffusion model that generates high-quality images reflecting the structure of a single image, without requiring any separate training.
A study revealed the mathematical correspondence between decision trees and diffusion models, unifying the two models.
The paradigm of visual AI is shifting from directly generating pixels to generating editable source code.
Ultra-lightweight 1-bit and Ternary Bonsai Image 4B models have been released, enabling high-quality image generation on local devices.
It proposes a new method that dramatically reduces memory usage by dividing the network into blocks and training them independently.
NVIDIA has released a Diffusion language model family that delivers inference speeds up to 6x faster than existing AR models.
MDLM improved agentic RL performance as a stronger text-based world model than AR.
NVIDIA released LongLive 2.0, an NVFP4 and parallelization-based system.
Orthrus, based on Qwen3, achieved up to 7.8x speedup through lossless parallel generation.
Orthrus, which uses a Diffusion module to boost token generation speed by up to 7.8x while maintaining the accuracy of existing Transformers, has been unveiled.
Microsoft is scouting alternative startups to reduce its dependence on OpenAI.
Scenema Audio has released weights for a zero-shot, emotion-decoupled voice generation model.
Scenema Audio has released the weights and inference code for a voice generation model that separates emotion and speaker identity.
By decomposing self-attention weights, editing directions for diffusion were found without any additional training.
NTM aims for both 4-step generation and exact likelihood by training the reverse step as a normalizing flow.
It explains how flow maps directly predict the integral of diffusion sampling.
Tortoise TTS V2 is a high-quality open-source TTS technology that utilizes an autoregressive and diffusion decoder.
NVIDIA Sol-RL reduces cost and accelerates convergence by separating FP4 rollouts from BF16 updates.
Modular Diffusers has been unveiled, allowing diffusion model pipelines to be assembled from reusable blocks.