Causal-rCM: A Unified Teacher-Forcing and Self-Forcing Open Recipe for Autoregressive Diffusion Distillation in Streaming Video Generation and Interactive World Models

arXiv 2026 59.5 method

TLDR

Extends rCM diffusion distillation to autoregressive video diffusion, achieving SOTA streaming video generation and interactive world models via teacher-forcing and self-forcing.

Reasoning

The paper presents a novel unified recipe combining teacher-forcing and self-forcing for diffusion distillation in autoregressive video generation, achieving state-of-the-art performance and 10x faster convergence. However, it relies solely on synthetic data, lacking real-world validation, and the abstract cuts off, leaving some details unclear.

Read-first score

Read-first score 59.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 54.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

95.3

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.95345351

Topical relevance 29%

77.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=experiment

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 504.

Keyword Scores

interactive world model

world model

video world model

generative world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

Deep Analysis

Innovations

Extending rCM, an advanced diffusion distillation framework, to autoregressive video diffusion.
Showing that teacher-forcing CM is the best complement to self-forcing DMD as an initialization strategy.
First implementation of teacher-forcing-based continuous-time consistency models (sCM/MeanFlow) for autoregressive video diffusion, enabled by a custom-mask FlashAttention-2 JVP kernel, achieving 10x faster convergence compared to discrete-time CMs.
Introducing Causal-rCM, a unified, scalable algorithm-infrastructure open recipe for diffusion distillation and causal training.
Achieving state-of-the-art streaming video generation performance in both frame-wise and chunk-wise settings using only synthetic data.
Applying Causal-rCM to Cosmos 3, an advanced omnimodal world foundation model, enabling an interactive world model.

Methodology

The paper extends the rCM diffusion distillation framework to autoregressive video diffusion, leveraging teacher-forcing (offline forward-divergence causal training) and self-forcing (on-policy reverse-divergence refinement). It implements continuous-time consistency models (sCM/MeanFlow) using a custom-mask FlashAttention-2 JVP kernel for efficient training. The method is evaluated on streaming video generation (frame-wise and chunk-wise) and action-conditioned world models, using only synthetic data for training.

Key Results

The distilled 2-step causal Wan2.1-1.3B model achieves a VBench-T2V score of 84.63 with only 1 or 2 sampling steps, achieving state-of-the-art streaming video generation performance.

Resources

Paper