epfl-vita/svi-benchmark

Stable Video Infinity (SVI) Benchmark Dataset

This benchmark dataset is introduced in the paper: [Stable Video Infinity: Infinite-Length Video Generation with Error Recycling](https://arxiv.org/abs/2510.09212) by Wuyang Li, Wentao Pan, Po-Chien Luan, Yang Gao, Alexandre Alahi (2025).

Project page: https://stable-video-infinity.github.io/homepage/ Code: https://github.com/vita-epfl/Stable-Video-Infinity

Abstract

We propose Stable Video Infinity (SVI) that is able to generate infinite-length videos with high temporal consistency, plausible scene transitions, and controllable streaming storylines. While existing long-video methods attempt to mitigate accumulated errors via handcrafted anti-drifting (e.g., modified noise scheduler, frame anchoring), they remain limited to single-prompt extrapolation, producing homogeneous scenes with repetitive motions. We identify that the fundamental challenge extends beyond error accumulation to a critical discrepancy between the training assumption (seeing clean data) and the test-time autoregressive reality (conditioning on self-generated, error-prone outputs). To bridge this hypothesis gap, SVI incorporates Error-Recycling Fine-Tuning, a new type of efficient training that recycles the Diffusion Transformer (DiT)'s self-generated errors into supervisory prompts, thereby encouraging DiT to actively identify and correct its own errors. This is achieved by injecting, collecting, and banking errors through closed-loop recycling, autoregressively learning from error-injected feedback. Specifically, we (i) inject historical errors made by DiT to intervene on clean inputs, simulating error-accumulated trajectories in flow matching; (ii) efficiently approximate predictions with one-step bidirectional integration and calculate errors with residuals; (iii) dynamically bank errors into replay memory across discretized timesteps, which are resampled for new input. SVI is able to scale videos from seconds to infinite durations with no additional inference cost, while remaining compatible with diverse conditions (e.g., audio, skeleton, and text streams). We evaluate SVI on three benchmarks, including consistent, creative, and conditional settings, thoroughly verifying its versatility and state-of-the-art role.

Dataset Overview

This repository hosts the benchmark datasets used for evaluating Stable Video Infinity, as well as pointers to the training data.

Benchmark Datasets

These datasets are generated by the Automatic Prompt Stream Engine (see Appendix. A of the paper for more details) and contain images and associated prompt streams according to specific storylines.

Training Datasets

The following are examples of training data formats or external datasets used for the SVI family:

Sample Usage

You can download the entire svi-benchmark dataset locally using the Hugging Face CLI:

huggingface-cli download --repo-type dataset vita-video-gen/svi-benchmark --local-dir ./data/svi-benchmark

Citation

If you find our work helpful for your research, please consider citing our paper:

@article{li2025stable,
      title={Stable Video Infinity: Infinite-Length Video Generation with Error Recycling}, 
      author={Wuyang Li and Wentao Pan and Po-Chien Luan and Yang Gao and Alexandre Alahi},
      journal={arXiv preprint arXiv: arXiv:2510.09212},
      year={2025},
}