notesum.ai

Published at November 22

Style-Friendly SNR Sampler for Style-Driven Generation

cs.CV

Released Date: November 22, 2024

Authors: Jooyoung Choi¹, Chaehun Shin¹, Yeongtak Oh¹, Heeseung Kim¹, Sungroh Yoon²

Aff.: ¹Data Science and AI Laboratory, ECE, Seoul National University; ²Data Science and AI Laboratory, ECE, Seoul National University and AIIS, ASRI, INMC, ISRC, and Interdisciplinary Program in AI, Seoul National University

Arxiv: http://arxiv.org/abs/2411.14793v1

Method	Model	Metrics
Method	Model	DINO $\uparrow$	CLIP-I $\uparrow$	CLIP-T $\uparrow$
Style-Aligned [14]	SDXL	0.410	0.675	0.340
RB-Mod [44]	Cascade	0.317	0.647	0.363
DCO [30]	SD3.5	0.399	0.661	0.355
SD3 sampler [8]	SD3.5	0.424	0.670	0.350
Style-friendly	SD3.5	0.489	0.698	0.349
IP-Adapter [61]	FLUX-dev	0.361	0.656	0.354
DCO [30]	FLUX-dev	0.373	0.643	0.353
SD3 sampler [8]	FLUX-dev	0.373	0.645	0.350
Style-friendly	FLUX-dev	0.461	0.686	0.344