AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Real-World Image Variation by Aligning Diffusion Inversion Chain

Yuechen ZHANG, Jinbo Xing, Eric Lo, Jiaya Jia

NeurIPS 2023 · Image generation › Text-to-image diffusion · approach: Inference-time T2I improvement · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
align the generation chain to the real-image inversion chain using RIVALStable Diffusion V1.5 image variation · Image variationPalette Distance: 3.650 ± 1.30 -> 1.674 ± 0.63Improved
add cross-image attention injection and attention fusion (ablation)RIVAL · Image variationImage Alignment: 0.760 -> 0.846Improved
add cross-image attention injection and attention fusion (ablation)RIVAL · Image variationPalette Distance: 2.576 -> 1.810Improved
add attention fusion (ablation)RIVAL with attention injection · Image variationImage Alignment: 0.818 -> 0.846Improved
add attention fusion (ablation)RIVAL with attention injection · Image variationPalette Distance: 1.902 -> 1.810Improved
add latent initialization alignment (ablation)RIVAL · Image variationImage Alignment: 0.653 -> 0.846Improved
add latent initialization alignment (ablation)RIVAL · Image variationPalette Distance: 3.741 -> 1.810Improved
add noise alignment (ablation)RIVAL · Image variationPalette Distance: 2.335 -> 1.810Improved
add noise alignment (ablation)RIVAL · Image variationImage Alignment: 0.839 -> 0.846Improved
remove cross-image self-attention injection (ablation)RIVAL · Image variationsemantic and exemplar correspondenceHurt
remove latent alignment and initialize from standard Gaussian noise (ablation)RIVAL · Image variationsemantic and tone consistencyHurt
use excessively long or short early alignment stages (ablation)RIVAL · Image variationvisual qualityHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy