Tag
This paper introduces Latent-to-4D, a method for direct 4D scene generation from video diffusion latents without retraining across generators, achieving better geometry and temporal stability than cascaded approaches.
MirrorWorld is a reflection-aware video inpainting framework that improves mirror reflection generation in videos by separately modeling semantic content (SRD) and geometric spatial arrangements (GTA), achieving better reflection reconstruction than existing image-based and video inpainting baselines.
Introduces UniWorld-View, a unified framework for large-baseline novel view synthesis from monocular inputs, integrating occlusion-aware point cloud rendering with video diffusion models for precise camera control and geometric consistency.
TenStrip/10Eros-Max is an experimental AI model that modifies the MiniMax H3 base model by transferring learned patterns from LTX 2.3, Wan 2.2, and Krea 2 video and image diffusion models using orthogonal projection, enhancing aesthetic and motion character while preserving core video and audio capabilities.
This paper proposes 'Mirror Learning', a framework for imitation learning from third-person observation that uses a fine-tuned video diffusion model for perspective transformation and an inverse dynamics model to synthesize pseudo first-person expert data, showing that this mirror data alone can train effective policies and improve behavior cloning.
DreamTraj predicts 6-DoF object trajectories from a single RGB image and a language instruction by decoding internal video diffusion latents, eliminating the need for video, depth, or CAD models at inference. It introduces the MOVEdataset with fine-grained language-to-motion annotations and achieves state-of-the-art performance while running 4.6x faster than generate-then-extract pipelines.
This paper introduces a rest-state framework that reconstructs articulated objects from a single closed configuration, using explicit meshes, vision-language outputs, and video diffusion models to generate and validate articulation hypotheses without observed motion.
WorldWeaver (W²) introduces cross-agent world state registers to multi-agent video diffusion models, enabling shared world state persistence across agents and views, improving logical consistency in two-agent Minecraft video generation.
Sol-Engine weekly update announces integration of Sol Attention, a training-free sparse attention method for video diffusion, with full paper coming next week.
AlayaWorld is a 15B-parameter interactive video world model that generates 24-fps video at 540p and 720p, using autoregressive latent chunk generation with bounded visual context and distillation to reduce inference steps. It achieves state-of-the-art performance on the iWorld-Bench for long-horizon generation.
Proposes StudioRecon, a method for 4D human-scene reconstruction from low-overlap camera captures, using separate priors for backgrounds (video diffusion) and humans (SMPL), with a recursive enhancement module. Accepted to SIGGRAPH 2026, outperforming prior methods by +1.5 to +5.0 dB PSNR.
Flex-Forcing introduces a unified framework for video diffusion that supports both autoregressive and bidirectional generation modes, offering flexible control for video generation tasks.
LongE2V leverages pre-trained video diffusion priors and introduces Autoregressive Unrolling, Adaptive Context Switching, Reencoding Alignment, and Event Voxel Density Augmentation to achieve high-quality video reconstruction, prediction, and frame interpolation from sparse event streams, outperforming state-of-the-art methods.
LingBot-Video is a 13B sparse-MoE video diffusion transformer (1.4B active) post-trained with RL as an action-conditioned world model, open-sourced with weights and code. It includes a physical-plausibility reward graded by a VLM and frames itself as a policy evaluator and action planner, though closed-loop robot results are absent.
MV-Forcing introduces a diffusion framework that combines temporal and view-wise autoregression to generate long, multi-view consistent videos of dynamic scenes, using a 4D geometric bridge and spatio-temporal distillation to enable arbitrary-length generation from a few-step student model.
ViDiHand leverages pretrained video diffusion model representations to reconstruct 4D hand motion directly from egocentric video frames, outperforming existing methods on ARCTIC, HOT3D, and HOI4D without detectors or optimization.
Enze Xie announces Sol Video Inference Engine, an agent-native, training-free full-stack accelerator for video diffusion that auto-tunes cache, sparse attention, token pruning, quantization, and kernel fusion, achieving >2× end-to-end speedup on large models like 64B Cosmos3-Super and 22B LTX-2.3.
FLAT is a method that directly decodes explicit triangle splats from compressed video diffusion latents in a single forward pass, improving geometric accuracy while enabling fast rasterization and physics-based interaction.
A scalable framework combines self-distillation and reinforcement learning to transfer task-solving abilities from vision-language models to video diffusion models without requiring labeled task-video data.
FadeMem introduces a distance-aware key-value memory consolidation mechanism that organizes historical video data into a temporal hierarchy, improving long-video generation under fixed cache constraints.