Autoregressive (AR) video models excel at causal generation, but their reliance on next-chunk prediction confines them to a short-sighted, reactive paradigm. This limitation is particularly consequential for reasoning-oriented generation, where achieving a target outcome through valid intermediate states matters more than local visual plausibility. To address this challenge, we propose Learning Prospective Reasoning with Autoregressive Video Models (ProAR), a novel framework that transforms autoregressive video generation into a goal-oriented reasoning process. ProAR introduces two key components: (1) To anchor generation to the long-range outcome, we integrate goal-frame prediction into the autoregressive loop via an asymmetric attention mask, enabling the predicted goal frame to guide the generation of intermediate states without being disrupted by them. (2) To guide short-range transitions, we introduce future representation self-alignment to encourage current hidden states to anticipate upcoming temporal dynamics. By leveraging teacher-forcing in AR training, we extract clean future representations in a single forward pass and align current representations with them using a lightweight, training-only predictor. Together, these two mechanisms seamlessly combine explicit, sparse target supervision with implicit, dense step-wise guidance, promoting coherent, goal-directed reasoning progress with modest computational cost. Experiments show that ProAR's complementary components consistently improve performance across diverse visual reasoning benchmarks. The framework proves highly training-efficient, surpassing fully trained standard AR baselines using only 25% of the training steps. This paradigm also demonstrates promising applicability to embodied reasoning tasks.
Figures & tables
Figure 1: Overview of ProAR. (a) Standard AR relies on local next-chunk prediction, leading to trajectory drift and task failure. (b) ProAR introduces prospective reasoning to anticipate immediate transitions and final outcomes, guiding generation toward task success.
Figure 2: Overview of ProAR. (a) Input: For a given video, the model processes the history, current, and goal streams within a forward pass. (b) Outcome: The causal DiT jointly denoises the current chunk and goal frame, where an attention mask enables the clean history and evolving goal belief to guide current generation. (c) Transition: A training-only projector aligns the middle-layer hidden states of the noisy current chunk (student) with the clean next chunk (teacher).
Figure 3: Qualitative Results. Across tasks from VBVR, VideoRLVR, and WorldArena, ProAR consistently improves reasoning performance by promoting goal completion and valid progression.
Figure 4: Performance curve on VBVR 10 tasks subsets.
Appendix figures & tables8 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 5: Failure Modes of Standard AR.
Figure 6: PCA Visualization of Future Representation Alignment.
Task
Description
Concentric Ring (G194)
Move two circles to the same center while preserving their sizes and colors.
Outline Match (G25)
Move each colored shape into its matching outline.
Stable Sort (G3)
Group shapes by type and sort each group from smallest to largest.
Star Match (G5)
Move each colored object to the star marker of the same color.
Domino Chain (O23)
Trigger the first domino and propagate the falling sequence through the complete chain.
Ball Eater (O31)
Guide the black ball to consume smaller balls, growing after each step.
Appendix
Table 8: Specifications of the 10 selected VBVR tasks ( Wang et al., 2026a ) .
Task
Description
Maze
Move the agent from the green start to the red goal through open cells without crossing any walls.
FlowFree
Connect each pair of same-colored endpoints with a continuous path, without overlapping other paths, until the entire grid is filled.
Sokoban
Move the agent through free cells and push the box onto the green target without crossing walls or pulling the box.
Appendix
Table 9: Specifications of the 3 VideoRLVR tasks ( Zhu et al., 2026b ) .
Autoregressive video generators synthesize long videos through chunk-wise rollouts, yet they are typically trained to predict only the immediate future. This limited prediction horizon makes it challenging to maintain video quality over extended rollouts. To address this problem, we introduce Prospective Forcing, a multi-head prediction framework that injects long-range predictive signals into autoregressive video representations. Built on a rolling-window generator, Prospective Forcing attaches lightweight, offset-specific heads that share the same conditioning information and independently predict chunks at different future temporal offsets. By extending supervision beyond the next chunk, these heads encourage the shared history representation to retain information useful for longer-horizon prediction, supporting temporally consistent generation. At inference time, we accelerate sampling by using accepted future-chunk predictions to skip full-window backbone executions. We equip the head with learnable embeddings to generate multiple candidates for the next chunk, and a learned verifier either selects a candidate or rejects all and falls back to standard autoregressive generation.
Yifan Shen, Yewei Xia, Gongxu Luo +6
Mohamed bin Zayed University of AI · Institute of Foundation Models · Carnegie Mellon University
Autoregressive video diffusion models generate videos chunk by chunk, yet producing plausible continuations does not guarantee that they capture how scenes evolve. Under teacher forcing, strong temporal redundancy may allow local visual continuity to serve as a shortcut for current-chunk denoising, reducing the pressure to model visual evolution. We introduce Future Chunk Prediction (FCP), an auxiliary objective jointly optimized with teacher-forced denoising that predicts temporally separated chunks without access to the intervening visual context. Drawing supervision directly from training videos, FCP requires no additional pretrained visual encoder and leaves autoregressive inference unchanged. Our method achieves superior physical plausibility over standard teacher-forced autoregressive training with a lower overall training compute budget, highlighting the effectiveness and efficiency of future-chunk supervision for modeling visual evolution. Intriguingly, our visualizations suggest that FCP fosters a more structured representation space, which may be valuable for building autoregressive world models.
Ruopeng Gao, Yukun Huang, Jiarui Hu +4
Nanjing University · ARC Lab, Tencent IEG · Zhejiang University
Autoregressive (AR) video diffusion models enable low-latency streaming generation by synthesizing videos chunk by chunk with cached visual context, but this chunk-wise formulation makes temporal instruction following ambiguous. A single global prompt does not specify which sub-event should be realized in each chunk, while naively switching to step-wise prompts often leads to delayed reactions, blended step semantics, and error propagation across prompt transitions. These failures are difficult to address with supervised fine-tuning or distillation alone: SFT suffers from exposure bias, while rollout-based distillation still optimizes low-level denoising or teacher-distribution matching rather than directly enforcing action ordering and prompt-transition correctness. We address these challenges with TempAct, a planner--executor reinforcement learning framework that jointly optimizes temporal decomposition and step-conditioned execution for temporally plausible AR video generation. TempAct uses an LLM planner to explore span-aware step prompts that are executable by the video model, and trains an AR diffusion executor to follow these prompts under its own generated histories. Its key mechanism is hierarchical group exploration: candidate plans form planning groups, and each plan induces an execution group of multiple continuations from a shared visual context, enabling plan-level credit assignment for long-horizon temporal outcomes and executor-level credit assignment for prompt-switch behavior. We further design hierarchical rewards that combine plan-quality and full-video temporal feedback for the planner with local transition-level step-following rewards, aesthetic regularization, and KL constraints for the executor. Experiments on Self-Forcing and LongLive show that TempAct improves temporal consistency while preserving overall visual quality.
Jing Wang, Xiangxin Zhou, Jiajun Liang +5
Shenzhen Campus of Sun Yat-Sen University · Tencent Hunyuan · Tsinghua University +2