# Distribution Matching Distillation

Duo v1 is a full-duplex world model that generates the assistant's audio and video in real time while continuously taking in the user's audiovisual input. Earlier world models largely kept intelligence outside the world, letting users steer scenes through keyboard, mouse or text prompts while nothing inside perceived the user or decided how to respond. Duo v1 cuts interaction into 17-frame chunks, about 0.71 s at 24 fps; a language intent module decides per chunk whether to speak or wait, and a causal audiovisual module generates the matching sound and picture. Of 180 checks across 30 real-time duplex cases and six binary criteria, 168 passed, including all 90 on language, speech intelligibility and visual identity, while content responsiveness reached 73.3%.

Duo v1 is a full-duplex world model that generates the assistant's audio and video in real time while continuously taking in the user's audiovisual input. Earlier world models largely kept intelligence outside the world, letting users steer scenes through keyboard, mouse or text prompts while nothing inside perceived the user or decided how to respond. Duo v1 cuts interaction into 17-frame chunks, about 0.71 s at 24 fps; a language intent module decides per chunk whether to speak or wait, and a causal audiovisual module generates the matching sound and picture. Of 180 checks across 30 real-time duplex cases and six binary criteria, 168 passed, including all 90 on language, speech intelligibility and visual identity, while content responsiveness reached 73.3%.

DuoMatching introduces a unified joint-marginal distribution matching framework: beyond existing joint DMD, it applies direct frame-level marginal supervision from an image generator, uses LatentBridge to resolve the video-image latent mismatch, and uses Latent Variation Sampling to spread frame-level supervision across temporal segments; in causal and bidirectional few-step video generation it improves visual quality, composition and semantic alignment while largely preserving motion dynamics, with human overall preference above 80% against all evaluated baselines.

DuoMatching introduces a unified joint-marginal distribution matching framework: beyond existing joint DMD, it applies direct frame-level marginal supervision from an image generator, uses LatentBridge to resolve the video-image latent mismatch, and uses Latent Variation Sampling to spread frame-level supervision across temporal segments; in causal and bidirectional few-step video generation it improves visual quality, composition and semantic alignment while largely preserving motion dynamics, with human overall preference above 80% against all evaluated baselines.

Memorizon proposes a training recipe in which each scored chunk retrieves its own top-k latents by camera co-visibility and their union forms a bounded shared memory bank, so a training sample can cover a span of any length while the attention sequence stays bounded; revisit consistency improves on every split over a sliding-window baseline, and a span long enough to reach the first visit adds further gains at some cost in image quality.

Memorizon proposes a training recipe in which each scored chunk retrieves its own top-k latents by camera co-visibility and their union forms a bounded shared memory bank, so a training sample can cover a span of any length while the attention sequence stays bounded; revisit consistency improves on every split over a sliding-window baseline, and a span long enough to reach the first visit adds further gains at some cost in image quality.

The work presents SoL-Refiner, a one-step video refiner that turns low-resolution generator outputs into 4K video through a three-stage recipe of high-resolution continual training, reinforcement-learning post-training with frame-based reward models, and final one-step distillation, and introduces Refiner-Bench, a video refinement benchmark using a shared-input protocol to compare refiners at roughly 2K output resolution; at 2K the one-step model outperforms all evaluated external refiners on VBench and UniPercept averages, at 3840×2176 it improves both metrics over the three-step LTX-2.3 Refiner, and with the complete acceleration stack it achieves an 8.91× speedup in refinement latency over the same baseline in the 2K latency setting.

The work presents SoL-Refiner, a one-step video refiner that turns low-resolution generator outputs into 4K video through a three-stage recipe of high-resolution continual training, reinforcement-learning post-training with frame-based reward models, and final one-step distillation, and introduces Refiner-Bench, a video refinement benchmark using a shared-input protocol to compare refiners at roughly 2K output resolution; at 2K the one-step model outperforms all evaluated external refiners on VBench and UniPercept averages, at 3840×2176 it improves both metrics over the three-step LTX-2.3 Refiner, and with the complete acceleration stack it achieves an 8.91× speedup in refinement latency over the same baseline in the 2K latency setting.

WorldLine introduces an action-driven visual simulator that decouples transferable robot-object dynamics learning from heterogeneous action grounding: it pretrains on more than 10,000 hours of action-free robot videos, grounds them with over 2,000 hours of action trajectories across more than ten embodiments through image-space action maps, adds multi-view, failure-enriched and relational-regularized training, and distills a robot-focused few-step causal rollout; across held-out and out-of-domain settings it maintains visual quality and robot-motion agreement, improves robot-mask IoU by 0.1626 over the strongest baseline on failed trajectories, predicts trajectory success with 74% mean accuracy on RoboTwin and AgiBot, and improves task success by up to 21.

WorldLine introduces an action-driven visual simulator that decouples transferable robot-object dynamics learning from heterogeneous action grounding: it pretrains on more than 10,000 hours of action-free robot videos, grounds them with over 2,000 hours of action trajectories across more than ten embodiments through image-space action maps, adds multi-view, failure-enriched and relational-regularized training, and distills a robot-focused few-step causal rollout; across held-out and out-of-domain settings it maintains visual quality and robot-motion agreement, improves robot-mask IoU by 0.1626 over the strongest baseline on failed trajectories, predicts trajectory success with 74% mean accuracy on RoboTwin and AgiBot, and improves task success by up to 21.

The work presents WorldCrafter, a camera-controllable autoregressive video world model that learns a camera-queryable implicit 3D-aware memory, compressing historical latent frames into a fixed budget of memory tokens read out under the requested viewpoint and injected into the video diffusion transformer before denoising, so that a single image or text prompt supports streaming, minute-scale scene exploration with better revisit consistency and camera-control accuracy than the evaluated baselines while preserving visual quality.

The work presents WorldCrafter, a camera-controllable autoregressive video world model that learns a camera-queryable implicit 3D-aware memory, compressing historical latent frames into a fixed budget of memory tokens read out under the requested viewpoint and injected into the video diffusion transformer before denoising, so that a single image or text prompt supports streaming, minute-scale scene exploration with better revisit consistency and camera-control accuracy than the evaluated baselines while preserving visual quality.