Records found: 10
'An accessible explainer of how modern AI systems combine latent diffusion and transformers to generate realistic video clips, why the outputs vary, and what the costs and benefits are.'
NVIDIA introduces GraspGen, a diffusion-based framework that significantly improves 6-DOF robotic grasping using large-scale synthetic data and innovative training, achieving superior performance in simulation and real-world tests.
New research reveals that generalization in flow matching models emerges from the neural network's approximation of velocity fields during early trajectories, challenging the belief that stochasticity drives this process.
Samsung Research developed ANSE, a novel framework that uses attention-based uncertainty estimation to improve noise seed selection in text-to-video diffusion models, boosting video quality and semantic consistency with minimal computational overhead.
Researchers at the National University of Singapore introduce Dimple, a novel discrete diffusion multimodal language model that enhances text generation efficiency and control, outperforming similar autoregressive models.
MMaDA is a novel unified multimodal diffusion model that excels in textual reasoning, visual understanding, and image generation, outperforming existing systems across multiple benchmarks.
New research uncovers how AI image models mistakenly include modern devices in historical scenes, revealing challenges in achieving historical accuracy in AI-generated images.
DanceGRPO introduces a unified reinforcement learning framework that enhances visual generation across multiple paradigms and tasks, significantly improving visual quality and alignment with human preferences.
MirrorVerse and MirrorFusion 2.0 introduce innovative datasets and training techniques that significantly enhance diffusion models' capacity to generate accurate and photorealistic mirror reflections.
Decoupled Diffusion Transformers (DDT) separate semantic encoding and detail decoding to speed up training and improve image generation quality, setting new benchmarks on ImageNet datasets.