aiqbench analysis
This paper introduces Conditional Residual Prediction (CRP), a training strategy that enables causal video diffusion models to match the quality of bidirectional counterparts without requiring large teacher models or complex distillation. By forcing the model to rely primarily on current inputs and treating historical context as a residual correction, CRP significantly reduces error propagation during autoregressive generation. The resulting Optica model demonstrates that high-fidelity long-video synthesis is achievable through simpler, more scalable architectures.
Conditional Residual Prediction: Improving Autoregressive Video Diffusion without a Bidirectional Teacher
arXiv:2610.11479v1 Announce Type: cross Abstract: Causal video diffusion models generate video autoregressively, which suits streaming, interactive, and long-video generation. Under standard training, however, they often yield lower generation quality than bidirectional models of the same size. Many existing approaches address this gap by initializing from or distilling a pretrained bidirectional teacher. We instead train a causal model from an image-model initialization, with no bidirectional video model at any stage. Because this path requires neither a large bidirectional teacher nor a complex distillation pipeline, it is simpler and more scalable. On this path, we find that a causal model trained on ground-truth history becomes strongly dependent on it, so that at inference errors in its own generated history propagate forward. We hypothesize that much of this dependence is unnecessary, because the current input already determines much of what the history provides. We propose Conditional Residual Prediction (CRP), a simple recipe for reducing a model's reliance on a condition: the model first predicts the target without the condition, and the condition may only add a residual on top of this prediction. Applied to history, CRP makes the model predict each chunk from the present as far as it can and use the past only for what the present cannot supply. In controlled experiments, CRP nearly closes the 6.14-point gap to a bidirectional model trained under the same setup. Scaling this recipe, we train Optica, a 2B-parameter causal video model that autoregressively generates 5-second 480p videos and reaches 82.78 on VBench with only about 15M training videos.
Abstract:Causal video diffusion models generate video autoregressively, which suits streaming, interactive, and long-video generation. Under standard training, however, they often yield lower generation quality than bidirectional models of the same size. Many existing approaches address this gap by initializing from or distilling a pretrained bidirectional teacher. We instead train a causal model from an image-model initialization, with no bidirectional video model at any stage. Because this path requires neither a large bidirectional teacher nor a complex distillation pipeline, it is simpler and more scalable. On this path, we find that a causal model trained on ground-truth history becomes strongly dependent on it, so that at inference errors in its own generated history propagate forward. We hypothesize that much of this dependence is unnecessary, because the current input already determines much of what the history provides. We propose Conditional Residual Prediction (CRP), a simple recipe for reducing a model's reliance on a condition: the model first predicts the target without the condition, and the condition may only add a residual on top of this prediction. Applied to history, CRP makes the model predict each chunk from the present as far as it can and use the past only for what the present cannot supply. In controlled experiments, CRP nearly closes the 6.14-point gap to a bidirectional model trained under the same setup. Scaling this recipe, we train Optica, a 2B-parameter causal video model that autoregressively generates 5-second 480p videos and reaches 82.78 on VBench with only about 15M training videos.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2610.11479 [cs.CV]
(or arXiv:2610.11479v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2610.11479 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 352 words
Related coverage
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026
Test-Time Noise Guided Adaptation for Realistic Autoregressive Video Generation
arXiv · Jul 20, 2026
OneFixer: High-Quality and Consistent One-Step Autoregressive 3DGS Refinement for Driving Scenes
arXiv · Sep 29, 2026
The Machines that Make the Machines
NVIDIA · Oct 7, 2026