Search:
Match:
1 results

Inference-based GAN for Long Video Generation

Published:Dec 25, 2025 20:14
1 min read
ArXiv

Analysis

This paper addresses the challenge of generating long, coherent videos using GANs. It proposes a novel VAE-GAN hybrid model and a Markov chain framework with a recall mechanism to overcome the limitations of existing video generation models in handling temporal scaling and maintaining consistency over long sequences. The core contribution lies in the memory-efficient approach to generate long videos with temporal continuity and dynamics.
Reference

Our approach leverages a Markov chain framework with a recall mechanism, where each state represents a short-length VAE-GAN video generator. This setup enables the sequential connection of generated video sub-sequences, maintaining temporal dependencies and resulting in meaningful long video sequences.