Diffusion distillation represents a highly promising direction for achieving
faithful text-to-image generation in a few sampling steps. However, despite
recent successes, existing distilled models still do not provide the full
spectrum of diffusion abilities, such as real image inversion, which enables
many precise image manipulation methods. This work aims to enrich distilled
text-to-image diffusion models with the ability to effectively encode real
images into their latent space. To this end, we introduce invertible
Consistency Distillation (iCD), a generalized consistency distillation
framework that facilitates both high-quality image synthesis and accurate image
encoding in only 3-4 inference steps. Though the inversion problem for
text-to-image diffusion models gets exacerbated by high classifier-free
guidance scales, we notice that dynamic guidance significantly reduces
reconstruction errors without noticeable degradation in generation performance.
As a result, we demonstrate that iCD equipped with dynamic guidance may serve
as a highly effective tool for zero-shot text-guided image editing, competing
with more expensive state-of-the-art alternatives.

通过应用可逆一致性蒸馏 (invertible Consistency Distillation, iCD) 框架，实现了在少于 4 个推理步骤中高质量图像合成和准确图像编码的目标，使得具有动态引导的 iCD 成为零样本文本引导图像编辑的高效工具。

用于文本引导图像编辑的可逆一致性蒸馏方法（约 7 步）

Invertible Consistency Distillation for Text-Guided Image Editing in  Around 7 Steps

The joint bilateral filter, which enables feature-preserving signal smoothing
according to the structural information from a guidance, has been applied for
various tasks in geometry processing. Existing methods either rely on a static
guidance that may be inconsistent with the input and lead to unsatisfactory
results, or a dynamic guidance that is automatically updated but sensitive to
noises and outliers. Inspired by recent advances in image filtering, we propose
a new geometry filtering technique called static/dynamic filter, which utilizes
both static and dynamic guidances to achieve state-of-the-art results. The
proposed filter is based on a nonlinear optimization that enforces smoothness
of the signal while preserving variations that correspond to features of
certain scales. We develop an efficient iterative solver for the problem, which
unifies existing filters that are based on static or dynamic guidances. The
filter can be applied to mesh face normals followed by vertex position update,
to achieve scale-aware and feature-preserving filtering of mesh geometry. It
also works well for other types of signals defined on mesh surfaces, such as
texture colors. Extensive experimental results demonstrate the effectiveness of
the proposed filter for various geometry processing applications such as mesh
denoising, geometry feature enhancement, and texture color filtering.

本文提出了一种新的静态 / 动态滤波器技术，通过利用静态和动态引导，实现了最先进的结果，适用于网格面法线，同时也适用于网格表面上定义的其他类型的信号，例如纹理颜色，能够对网格几何进行尺度感知和特征保留的过滤。