Online reinforcement learning (RL) fine-tunes diffusion models with black-box rewards, but reverse-process methods such …

模仿学习的老问题是「多峰动作分布」——两种都对的做法被回归平均成一种错的。Diffusion Policy 用条件去噪扩散直接建模动作分布，在 12 个任务、4 个基准上平均成功率提升 46.9%，此后 DP3、DPPO 相继跟进，π0 的流匹配与 GR00T 的扩散 Transformer 把它推成了 VLA 时代的标配动作头。

Diffusion large language models (DLLMs) generate text through iterative block denoising, and multi-branch speculative de…

Sparse attention is a primary approach to reducing the latency of diffusion transformers in long-sequence generation tas…

Bidirectional video diffusion models denoise entire videos in parallel, yet when trained on effectively unlimited in-dis…

Multimodal Diffusion Transformers (MM-DiTs) jointly process visual and textual representations throughout generation. Th…

Autoregressive video diffusion supports streaming generation and interactive control, but its KV cache grows continuousl…

Masked diffusion models (MDMs) generate sequences by progressively unmasking several tokens per denoising step, but thei…

Diffusion Transformers (DiTs) have established themselves as a scalable backbone for high-fidelity image synthesis. Howe…

Few-step autoregressive video diffusion generates a long video by splitting the video into temporal chunks and generatin…

Diffusion Large Language Models (dLLMs) have recently emerged as a promising alternative to autoregressive LLMs by enabl…

Despite impressive visual quality, state-of-the-art video diffusion models often generate content that violates real-wor…

3D diffusion policies are strong at generating geometrically grounded actions from current observations, but successful …

Nunchux AI has released VC-Attention, a training-free low-bit attention kernel built for video Diffusion Transformers (D…

Masked diffusion language models (dLLMs) generate text by iteratively denoising masked tokens with bidirectional attenti…

Diffusion large language models (dLLMs) achieve high decoding efficiency through block-parallel, arbitrary-order generat…

Autoregressive (AR) video diffusion models have shown great potential in real-time video generation. Recent methods dist…

Visual representations are largely developed for understanding, yet what makes them useful for generation remains unclea…

Most image generation models rely on uniform tokenization, allocating the exact same computational budget to equally-siz…

Large-scale Digital Surface Models (DSMs) can be produced cost-effectively from satellite images via stereo-photogrammet…

Recent years have witnessed major progress in joint audio-video generation. Existing models still suffer from limited pe…

Language models can be adapted by changing the computations applied to individual tokens. Quantum circuits offer one suc…

High-quality texture generation is essential for creating realistic and production-ready 3D assets. Recent multi-view di…

Google Research has introduced Retrieve-for-Train (R4T), a framework for search that returns coherent, diverse result se…

Visual goal and dynamics prediction can provide language-conditioned robot policies with both a target outcome and a rep…

3D point-cloud observations are inherently ambiguous in complex, cluttered manipulation scenes, where target objects may…

Monocular depth estimation is a ubiquitous yet highly ill-posed computer vision task, with downstream applications in sc…

Large Language Models (LLMs) owe much of their success to next-token prediction (NTP), but their autoregressive (AR) str…

Diffusion/flow models have emerged as the de facto paradigms for image generation, giving rise to a diverse family of mo…

Diffusion and autoregression (AR) have long been seen as different categories of generative models, with diffusion speci…

从 Peebles 与谢赛宁的 DiT 论文到 Sora 的时空 patch，讲清 Diffusion Transformer 的三个关键设计：patch 化、adaLN-Zero 条件注入与以计算量为标尺的可扩展性。

直接让网络输出一张合理的图像为什么难？扩散模型把「一步生成」反转成「多步去噪」：前向加噪提供训练素材，反向网络一步步剥离噪声，文本经 cross-attention 指挥去噪方向。本文讲清这套机制的完整逻辑，并对比扩散与自回归两条路线。

Representation autoencoders (RAEs) reuse features from a pretrained visual encoder as reconstruction and diffusion laten…

Few-step autoregressive (AR) video diffusion enables low-latency streaming generation, but existing post-training method…

Representation Autoencoders (RAEs) enable diffusion models to operate in the feature spaces of pretrained visual encoder…

Alibaba's Qwen team has released Qwen-Image-2.1, a 7B diffusion transformer that handles text-to-image generation, multi…

In this paper, we propose SVEET, a framework that requires merely training on a pretrained bidirectional video diffusion…

Text-guided image editing must introduce the requested changes while preserving unrelated source content. Diffusion-base…

Video diffusion models repeatedly process long spatiotemporal token sequences during denoising, making attention a major…

Diffusion Transformers deliver state-of-the-art video generation, but their long spatiotemporal sequences make attention…

Orthrus is a hybrid autoregressive-diffusion architecture that accelerates autoregressive language-model inference by ge…

Multimodal agents can create complex videos in software such as Blender by coding without relying on diffusion models. Y…

Video diffusion models are stochastic and hard to control: precise content often requires repeated sampling without guar…

Diffusion-based models enable monocular geometry estimation, yet their pixel-space precision is limited by a shared, und…

Audio-video diffusion models rely on cross-modal attention to coordinate text, sound, and visual content, yet this same …

Tomislav Bezmalinović Oct 11, 2026 Odyssey AI Key Points California-based AI company Odyssey has launched a public resea…

Speculative decoding accelerates large language model inference by using a lightweight drafter to propose multiple token…

PagedAttention 解决了「KV 内存怎么分页」，但没解决「相同前缀为什么要重算」——SGLang 用 RadixAttention 把前缀复用变成运行时自动机制，再靠零开销调度、缓存感知路由与三级 KV 缓存一路打进了 xAI 与 Azure 的生产环境。本文拆解它的架构主线、2026 年的双周发版节奏、与 vLLM 的真实差距，以及上手要点。

作者从头预训练了3B参数的像素空间文生图Transformer Iris-3B，并将预训练的潜在空间模型FLUX.2 Klein base 4B转换为像素空间，在单目深度估计和图像修复/超分辨率任务上微调，发现像素空间生成先验并未带来显著提升。

论文提出生成感知潜空间压缩方法 GRACE，旨在解决高压缩率视频自编码器训练困难及压缩潜空间与预训练 DiT 分布不匹配的问题，从而加速视频扩散模型推理。