DeCoPrune: Training-Free Pruning Speeds Video Generation 4x

XingangP's team introduced DeCoPrune, a training-free token pruning method for autoregressive video diffusion models that uses denoising difficulty to decide which tokens to keep. It cuts historical tokens by 85% and speeds up video continuation by about 4.14x.

2026-10-12 ~ 2026-10-12 · 2 related posts

• DeCoPrune: training-free pruning cuts video diffusion memory tokens 85%, 4.14x faster — XingangP · 2026-10-12

1 near-duplicate retellings: XingangP