Vision & Video

ShallowStream uses shallow MLLM layers to encode incoming frames and maintain a lightweight retrieval index, reserving deeper computation for question answering. The paper reports up to 52.1× lower per-frame prefill latency and 11.9× lower end-to-end latency over 10 seconds.

A new arXiv paper proposes DM-Align, a single-stage framework that combines video generation distillation with human-preference alignment. It builds a preference-guided gradient from sample distribution gaps, aiming to reduce the cost and instability of conventional reinforcement-learning pipelines.

A controlled study separates frame selection, spatial compression, and budget reinvestment in long-video multimodal models. It finds that choosing the right frames matters most, while compression pays off only when its savings are used to inspect more moments.

Google DeepMind has introduced Agentic Video Understanding for Gemini, allowing the model to dynamically search and revisit relevant parts of a video across frames, audio and transcripts. Google reports up to 88% lower token usage, 66% lower cost and 7% higher accuracy on selected benchmarks.

Luce introduces a multimodal Gaussian representation that jointly models geometry and physically based materials for single-image-to-3D generation. Its decoder produces relightable PBR Gaussians and can optionally output a textured mesh with a tangent-space normal map.

Amap has released ABot-Recon, a streaming 3D reconstruction model that uses only the latest 12 frames as local context while reconstructing scenes spanning tens of thousands of frames. Its design targets drift, slowing inference, and growing memory usage in long video sequences.

Google DeepMind’s Gemini Omni 1.1 Flash adds scene extension, first-and-last-frame generation, low-resolution previews, video references, and upscaling to 4K.

SparsePR is a training-free sparse attention method for video generation and world models. It combines response-coupled partitioning with probe-fitted residual reconstruction to reduce executed attention pairs while preserving output quality.

InfinityEdit targets open-ended video streams rather than fixed clips, generating future segments that continue the scene while applying incoming edit instructions. Its lightweight adapter connects visual history, causal time flow, and edit prompts to a streaming video generator.

4DAnyone turns an uncalibrated monocular human video into reconstruction-ready, multiview-consistent videos and then lifts them into a 4D Gaussian Splatting representation. Its main contribution is a pair of context-management mechanisms designed to keep many generated views structurally and visually aligned.

Alibaba’s Wan 3.0 video generation model has entered public beta. Its most notable upgrade is support for document inputs such as doc, ppt, pdf, and md, pushing video generation closer to a productivity workflow.

JoyAI-Video-Edit targets a hard corner of video generation: editing in real time without future frames or a fixed video length. The paper combines autoregressive diffusion with distillation strategies to preserve source fidelity and long-range consistency.

InfiniSplat targets large-baseline novel view synthesis from a single image by replacing fixed pixel-aligned Gaussian prediction with geometry-guided support sampling and implicit Gaussian decoding.

Motion Beyond Morphology (MBM) reframes video motion transfer as a problem of preserving transferable dynamics rather than matching fixed structures. The framework aims to animate targets whose morphology may differ substantially from the reference video.

A new paper introduces 3DarkFusion, a method that fuses extremely noisy RGB observations with near-infrared cues in 3D space. The approach aims to recover clean RGB images without relying on clean RGB supervision.

Meshy T2 proposes a faster alternative to autoregressive mesh decoding by generating native polygon meshes with flow matching and per-vertex continuous latent tokens.

Pippa is pitching itself as a more ethical AI video platform by paying artists when users generate work in their styles. But its approach still depends partly on open models trained on broadly scraped online content.

VideoCoCo introduces a dual-engine framework that converts text prompts into executable Blender programs before transforming simulated drafts into photorealistic videos. The goal is to improve physical consistency in text-to-video generation.

O-VAD is a training-free framework for industrial video anomaly detection that tracks objects through time and reasons over their state trajectories. Its central claim is that frontier VLMs often fail in industrial settings because they are not given object-level evidence.

Netflix’s ID-V2V targets identity-preserving video restylization: propagate scene, lighting, and style edits from keyframes across a source video without losing facial likeness, expressions, gaze, or lip sync.

The paper tackles a practical weakness in long-horizon generative rendering: when a camera returns to a previously seen place, an autoregressive video model may redraw it differently. The proposed method restores historical latent chunks and uses 3D correspondences to guide attention, without post-training.