Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2509.10441 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911151529394176 |
|---|---|
| author | Han, Tao Xu, Wanghan Gong, Junchao Yue, Xiaoyu Guo, Song Zhou, Luping Bai, Lei |
| author_facet | Han, Tao Xu, Wanghan Gong, Junchao Yue, Xiaoyu Guo, Song Zhou, Luping Bai, Lei |
| contents | Arbitrary resolution image generation provides a consistent visual experience across devices, having extensive applications for producers and consumers. Current diffusion models increase computational demand quadratically with resolution, causing 4K image generation delays over 100 seconds. To solve this, we explore the second generation upon the latent diffusion models, where the fixed latent generated by diffusion models is regarded as the content representation and we propose to decode arbitrary resolution images with a compact generated latent using a one-step generator. Thus, we present the \textbf{InfGen}, replacing the VAE decoder with the new generator, for generating images at any resolution from a fixed-size latent without retraining the diffusion models, which simplifies the process, reducing computational complexity and can be applied to any model using the same latent space. Experiments show InfGen is capable of improving many models into the arbitrary high-resolution era while cutting 4K image generation time to under 10 seconds. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_10441 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | InfGen: A Resolution-Agnostic Paradigm for Scalable Image Synthesis Han, Tao Xu, Wanghan Gong, Junchao Yue, Xiaoyu Guo, Song Zhou, Luping Bai, Lei Computer Vision and Pattern Recognition Arbitrary resolution image generation provides a consistent visual experience across devices, having extensive applications for producers and consumers. Current diffusion models increase computational demand quadratically with resolution, causing 4K image generation delays over 100 seconds. To solve this, we explore the second generation upon the latent diffusion models, where the fixed latent generated by diffusion models is regarded as the content representation and we propose to decode arbitrary resolution images with a compact generated latent using a one-step generator. Thus, we present the \textbf{InfGen}, replacing the VAE decoder with the new generator, for generating images at any resolution from a fixed-size latent without retraining the diffusion models, which simplifies the process, reducing computational complexity and can be applied to any model using the same latent space. Experiments show InfGen is capable of improving many models into the arbitrary high-resolution era while cutting 4K image generation time to under 10 seconds. |
| title | InfGen: A Resolution-Agnostic Paradigm for Scalable Image Synthesis |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2509.10441 |