Saved in:
Bibliographic Details
Main Authors: Park, Jihun, Gim, Jongmin, Lee, Kyoungmin, Oh, Minseok, Choi, Minwoo, Kim, Jaeyeul, Park, Woo Chool, Im, Sunghoon
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.06144
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914167711072256
author Park, Jihun
Gim, Jongmin
Lee, Kyoungmin
Oh, Minseok
Choi, Minwoo
Kim, Jaeyeul
Park, Woo Chool
Im, Sunghoon
author_facet Park, Jihun
Gim, Jongmin
Lee, Kyoungmin
Oh, Minseok
Choi, Minwoo
Kim, Jaeyeul
Park, Woo Chool
Im, Sunghoon
contents We present a training-free style-aligned image generation method that leverages a scale-wise autoregressive model. While large-scale text-to-image (T2I) models, particularly diffusion-based methods, have demonstrated impressive generation quality, they often suffer from style misalignment across generated image sets and slow inference speeds, limiting their practical usability. To address these issues, we propose three key components: initial feature replacement to ensure consistent background appearance, pivotal feature interpolation to align object placement, and dynamic style injection, which reinforces style consistency using a schedule function. Unlike previous methods requiring fine-tuning or additional training, our approach maintains fast inference while preserving individual content details. Extensive experiments show that our method achieves generation quality comparable to competing approaches, significantly improves style alignment, and delivers inference speeds over six times faster than the fastest model.
format Preprint
id arxiv_https___arxiv_org_abs_2504_06144
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Training-Free Style-aligned Image Generation with Scale-wise Autoregressive Model
Park, Jihun
Gim, Jongmin
Lee, Kyoungmin
Oh, Minseok
Choi, Minwoo
Kim, Jaeyeul
Park, Woo Chool
Im, Sunghoon
Computer Vision and Pattern Recognition
We present a training-free style-aligned image generation method that leverages a scale-wise autoregressive model. While large-scale text-to-image (T2I) models, particularly diffusion-based methods, have demonstrated impressive generation quality, they often suffer from style misalignment across generated image sets and slow inference speeds, limiting their practical usability. To address these issues, we propose three key components: initial feature replacement to ensure consistent background appearance, pivotal feature interpolation to align object placement, and dynamic style injection, which reinforces style consistency using a schedule function. Unlike previous methods requiring fine-tuning or additional training, our approach maintains fast inference while preserving individual content details. Extensive experiments show that our method achieves generation quality comparable to competing approaches, significantly improves style alignment, and delivers inference speeds over six times faster than the fastest model.
title A Training-Free Style-aligned Image Generation with Scale-wise Autoregressive Model
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.06144