Text-based diffusion models have made signiffcant breakthroughs
in generating high-quality images and videos from
textual descriptions. However, the lengthy sampling time
of the denoising process remains a signiffcant bottleneck in
practical applications. Previous methods either ignore the
statistical relationships between adjacent steps or rely on
attention or feature similarity between them, which often
only works with specifc network structures. To address this
issue, we want to propose a novel training-free method which
both enhances the efficiency of diffusion models and minimizes
additional resources.
