#video
- 28 Oct 2026 · Paper summary
Paper summary: ChronoReplay: One Answer Interface Is Not Enough for Streaming Spatial Intelligence
Training a spatial VLM on one kind of answer (e.g. left/right) doesn't teach related ones (e.g. distance), which the authors call "interface lock-in"; ChronoReplay replays mixed spatial supervision as time-ordered episodes and improves three streaming benchmarks together.
- 16 Oct 2026 · Paper summary
Paper summary: OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding
OmniCompass is a manually annotated benchmark for video grounding and understanding, with a vision track and an audio-visual track (when, where and what); across 50 MLLMs, the best scores are only around 40%.
- 27 Aug 2026 · Paper summary
Paper summary: VGI-Bench: Probing Visual Intelligence in Video Generation Models
VGI-Bench tests visual reasoning in video generation models with 27 tasks and 810 instances that require a valid process, not just a plausible final frame; the best model, Seedance 2.0, scores 51.0%.
- 17 Aug 2026 · Paper summary
Paper summary: Can We Defend Against AI-Generated Video Attacks on Real-World Crisis Events? A Systematic Evaluation of Detectors, Generators and Social Dissemination
RA-Bench pairs 1,830 real crisis videos with 16,056 generated clips from nine generators, and finds that no family of AI-video detectors generalises consistently, including when videos are re-shared on social media.
- 22 Jul 2026 · Paper summary
Paper summary: ABot-World-0: Infinite Interactive World Rollout on a Single Desktop GPU
ABot-World-0 is an action-conditioned video world model for long, closed-loop interaction that streams 720p at up to 16 FPS on a single RTX 5090, with LongForcing to limit drift over long rollouts.
- 10 Jul 2026 · Paper summary
Paper summary: Vidu S1: A Real-Time Interactive Video Generation Model
Vidu S1 generates infinite-length interactive video of digital characters in real time, controlled by voice at any moment, at 540p and up to 42 FPS on consumer GPUs.
- 28 May 2026 · Paper summary
Paper summary: Gamma-World: Generative Multi-Agent World Modeling Beyond Two Players
Gamma-World is a generative world model for interactive video with many simultaneously controlled agents, using a parameter-free simplex rotary agent encoding and hub-token attention, distilled for real-time 24 FPS generation.
- 16 Apr 2026 · Paper summary
Paper summary: Seedance 2.0: Advancing Video Generation for World Complexity
Seedance 2.0 is ByteDance's joint audio–video generation model accepting text, image, audio and video references, generating 4–15 second clips at 480p or 720p.
- 8 Apr 2026 · Paper summary
Paper summary: Video-MME-v2: Towards the Next Stage in Benchmarks for Comprehensive Video Understanding
Video-MME-v2 is a harder, human-built video understanding benchmark with a three-level hierarchy and group-based scoring that only credits consistent, well-reasoned answers; the best model, Gemini-3-Pro, still trails human experts substantially.
- 24 Feb 2026 · Paper summary
Paper summary: A Very Big Video Reasoning Suite
VBVR is a video-reasoning dataset of 200 curated tasks and over a million clips, about 1,000× larger than earlier ones, with a rule-based benchmark used for an early study of how video reasoning scales.
- 19 Dec 2025 · Paper summary
Paper summary: Kling-Omni Technical Report
Kling-Omni is one end-to-end framework for video generation, editing and reasoning-based editing from mixed inputs (text, reference images, video context), built on a dedicated data system.
- 10 Dec 2025 · Paper summary
Paper summary: Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance
Wan-Move adds precise motion control to an image-to-video model by turning dense point trajectories into a motion-aware latent condition, with no architecture change; user studies rate it close to Kling 1.5 Pro's Motion Brush.
- 20 Nov 2025 · Paper summary
Paper summary: Kandinsky 5.0: A Family of Foundation Models for Image and Video Generation
Kandinsky 5.0 is an open family of image and 10-second video generation models (6B Image Lite, 2B Video Lite, 19B Video Pro), with the full data-curation and multi-stage training pipeline described.
- 7 Nov 2025 · Paper summary
Paper summary: Thinking with Video: Video Generation as a Promising Multimodal Reasoning Paradigm
"Thinking with Video" uses a video generation model (Sora-2) as a reasoner, drawing its solution as video frames; on a new benchmark it rivals strong VLMs on visual puzzles and reaches 92% on MATH.
- 18 Oct 2024 · Paper summary
Paper summary: Movie Gen: A Cast of Media Foundation Models
Meta's Movie Gen is a set of foundation models for 1080p HD video with synchronised audio, plus personalisation and instruction-based editing; the largest video model is a 30B Transformer.
- 5 Sept 2024 · Paper summary
Paper summary: Loopy: Taming Audio-Driven Portrait Avatar with Long-Term Motion Dependency
Loopy is a video diffusion model that animates a portrait from audio alone, using long-term motion modules instead of the spatial motion templates other methods add to keep movement stable.
- 2 Aug 2024 · Paper summary
Paper summary: SAM 2: Segment Anything in Images and Videos
SAM 2 extends Segment Anything to video with a streaming-memory Transformer and a data engine that built the largest video segmentation dataset to date.
- 7 Jun 2024 · Paper summary
Paper summary: ShareGPT4Video: Improving Video Understanding and Generation with Better Captions
ShareGPT4Video builds dense, precise video captions (40K with GPT-4V, then 4.8M with its own captioner) to improve both video understanding models and text-to-video generation.
- 21 May 2024 · Paper summary
Paper summary: FIFO-Diffusion: Generating Infinite Videos from Text without Training
FIFO-Diffusion generates arbitrarily long videos from a pretrained text-to-video diffusion model, with no extra training, by denoising a queue of frames at increasing noise levels.
- 24 Jan 2024 · Paper summary
Paper summary: Lumiere: A Space-Time Diffusion Model for Video Generation
Lumiere generates the full duration of a video in one pass with a Space-Time U-Net that downsamples in both space and time, instead of generating keyframes and then filling in between.