Category

Can MiniMax-H3 Reason About the Physical World? An Evaluation of Omni-Modal Generative Model

Evaluates whether MiniMax-H3 can infer latent physical events from incomplete multimodal evidence and produce videos consistent with those inferences. Builds a 517-instance benchmark across four reasoning scenarios (multi-view, audio disambiguation, video continuation, audiovisual integration) and quantifies modality-specific failure modes.

RewardVerse: Rubric-Guided Policy Optimization for Video Reward Modeling

Introduces a rubric-based approach for video reward modeling that generates explicit, query-adaptive evaluation criteria before scoring to reduce scalar drift; proposes RGPO, a two-stage training (seed warm-up + joint optimization) that achieves strong pointwise and pairwise evaluation with high data efficiency.

WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory

Learns a camera-queryable implicit 3D-aware memory that compresses multi-view history into target-view tokens to enable long-horizon, camera-controllable video generation. Improves revisit consistency and camera-control accuracy and supports streaming exploration from a single image or text prompt.

Training Object Permanence in World Models

Provides WROP: a 1.5M-sample synthetic video corpus and a 300-question exam for training and evaluating object permanence and solidity in video world models. Includes 150 Blender task generators, a human Elo benchmark across 14 models, and a fine-tuned 16B continuation model (PWM-WROP).

The Past Frames the Future: Memory for Autoregressive Video Generation

Surveys memory mechanisms for autoregressive video generation, framing memory as persistent historical information that influences future generation. Organizes work by Forms, Functions, Operations, Learning, and Evaluation, and synthesizes challenges for long-horizon consistency and memory-aware learning.

WanPE: Towards Cinematic Prompt Enhancement for Modern Text-to-Video Generation

Transforms user prompts into shot-level cinematic directions for text-to-video generation, using a 397B prompt-enhancer trained on 1.05M videos and SC-GRPO to preserve semantic consistency across shots; evaluated on WanPEval (5–30s) with large human-preference gains.

Adaptive Reward Routing: Dynamic Multi-Reward Optimization for Joint Audio-Video Diffusion via Forward-Process RL

Proposes a forward-process RL method that dynamically localizes gradient updates and adapts multi-reward coordination for joint audio–video diffusion models. Uses bidirectional cross-attention responses for token/layer routing and preference-preserving reweighting to balance competing objectives, improving modality quality, alignment, and synchronization.

Beyond the Timeline: Augmenting Long-Video Memory with Grounded Entity Biographies

Groups visually grounded appearances of the same physical instance into persistent, retrievable “biographies” so agents can follow objects across hours or days for long-video question answering. Links identity-aware observations to episodic context and visual evidence; improves EgoLifeQA to 72.0% and increases evidence-window reach from 37.6% to 58.9%.

FrameMorrow: Future-guided Frame Selection with Prospective Tokens for Long-Horizon Video Generation

Predicts compact 'prospective tokens' that summarize upcoming information needs and uses them to select a small set of past frames for conditioning long-horizon video generation, improving long-range consistency, visual quality, and action alignment while remaining plug-and-play across diverse generators.

Hear the World in Stereo: Learning Dynamic Spatial Correspondence for Immersive Joint Video-Audio Generation

Generates stereo audio whose perceived locations evolve consistently with visible source motion in joint video–audio generation. Introduces StereoBind (motion-conditioned stereo synthesis), a large-scale training set StereoWorld-29K, and an evaluation suite StereoWorldBench for dynamic spatial correspondence.

OneStreamer: Unifying Perception, Memory, and Proactive Response in Streaming Video Interaction

Turns live video into reusable textual memory and timely responses by training a streaming video LLM to proactively generate time-grounded captions and event summaries. Key components: Proactive Hierarchical Caption Memory (PHCM) for multi-scale records and Proactive State Transition Learning (PSTL) to balance response timing; trained on the OneStreamer-1M streaming dataset.

Rethinking Long-Video Efficiency: A Joint Allocation Perspective on Frames, Pixels, and Front-End Latency

Introduces LoHi, a training-free, single-pass method that mixes dense low-resolution video streams with sparse high-resolution frames to improve long-video vision-language model accuracy under strict token budgets while cutting front-end decoding latency.