GRACE：面向生成感知的潜在压缩加速视频生成

AI 摘要

KAIST AI 与 Kakao Corp. 的研究者提出 GRACE，一种两阶段框架，可在保持与预训练 DiT 兼容的前提下压缩视频自编码器：保留冻结的基础 latent，学习残差 latent 补偿强压缩丢失的信息，并在冻结 DiT 的特征空间中对齐压缩 latent，使自编码器面向生成而非仅重建优化。随后仅需轻量微调 DiT 并采用非对称去噪（基础先于残差去噪）。该方法将 Wan2.1-I2V-14B 在 480x832x81 下的 token 数减少 8 倍、延迟降低 11.1 倍，同时在 VBench 上匹配压缩前管线的生成质量。

正文节选

GRACE: Generation-aware latent compression for efficient video generation Abstract Highly compressed video autoencoders offer an effective way to accelerate video diffusion models, as the Diffusion Transformer (DiT) operates on far fewer tokens. However, such autoencoders are challenging to train, since a higher compression ratio degrades reconstruction quality and recovering it requires more channels, which is known to slow the convergence of the DiT. The compressed latent also differs from the

发布时间：—

抓取时间：2026-10-08 12:30

来源机构：Hugging Face