PreCoG: Preview-Edit for Efficient Video Correction in Generation-Time

Key points

• Iterative text-to-video generation relies on feedback to correct compositional and temporal failures.1
• PreQA diagnoses failures from low-cost previews and constructs corrective conditions, while PreEdit edits cached generation trajectories under these conditions, attenuating orthogonal changes to limit disruption of already correct content.1
• PreCoG completes generation only for the selected candidate, reusing its cached trajectory to reduce redundant computation and help preserve semantic consistency between the preview and the final video.1
• Experiments show that PreCoG achieves the highest VBench and T2V-CompBench scores and approaches VSR's leading VS2 score at $0.56\times$ its inference time.1

Sources (1)

• [1]PreCoG: Preview-Edit for Efficient Video Correction in Generation-Time
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 08:01 AM

  We introduce PreCoG, an iterative framework for correcting semantic failures during video generation.

  Iterative text-to-video generation relies on feedback to correct compositional and temporal failures.

Extractive summary: sentences quoted from the sources.