Decoupled Latent Diffusion Model for Enhancing Image Generation

Latent Diffusion Models have emerged as an efficient alternative to conventional diffusion approaches by compressing high-dimensional images into a lower-dimensional latent space using a Variational Autoencoder (VAE) and performing diffusion in that space. In standard Latent Diffusion Model (LDM), t...

Full description

Saved in:
Bibliographic Details
Main Authors: Hyun-Tae Choi, Kensuke Nakamura, Byung-Woo Hong
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11091282/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849245913476759552
author Hyun-Tae Choi
Kensuke Nakamura
Byung-Woo Hong
author_facet Hyun-Tae Choi
Kensuke Nakamura
Byung-Woo Hong
author_sort Hyun-Tae Choi
collection DOAJ
description Latent Diffusion Models have emerged as an efficient alternative to conventional diffusion approaches by compressing high-dimensional images into a lower-dimensional latent space using a Variational Autoencoder (VAE) and performing diffusion in that space. In standard Latent Diffusion Model (LDM), the latent code is formed by sampling from a Gaussian distribution (i.e., combining both the mean and the standard deviation), which helps regularize the latent space but appears to contribute little beyond the deterministic component. Motivated by recent empirical observations that the decoder relies primarily on the latent mean, our work reexamines this paradigm and proposes a decoupled latent diffusion model that focuses on a simplified latent representation. Specifically, we compare three configurations: (i) the standard latent code, (ii) a concatenated representation that explicitly preserves both mean and variance, and (iii) a deterministic mean-only representation. Our extensive experiments on multiple benchmark datasets demonstrate that, when compared to the standard approach, the mean-only configuration not only maintains but in many cases improves synthesis quality by producing sharper and more coherent images while reducing unnecessary noise. These findings suggest that a simplified, deterministic latent representation can yield more stable and efficient generative models, challenging the conventional reliance on latent sampling in diffusion-based image synthesis.
format Article
id doaj-art-0754ed6ec3f041b4abd4a58ebccf0d34
institution Kabale University
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-0754ed6ec3f041b4abd4a58ebccf0d342025-08-20T03:58:40ZengIEEEIEEE Access2169-35362025-01-011313050513051610.1109/ACCESS.2025.359216311091282Decoupled Latent Diffusion Model for Enhancing Image GenerationHyun-Tae Choi0https://orcid.org/0000-0001-8268-0705Kensuke Nakamura1https://orcid.org/0000-0002-6858-3551Byung-Woo Hong2https://orcid.org/0000-0003-2752-3939Department of Artificial Intelligence, Chung-Ang University, Seoul, South KoreaDepartment of Artificial Intelligence, Chung-Ang University, Seoul, South KoreaDepartment of Artificial Intelligence, Chung-Ang University, Seoul, South KoreaLatent Diffusion Models have emerged as an efficient alternative to conventional diffusion approaches by compressing high-dimensional images into a lower-dimensional latent space using a Variational Autoencoder (VAE) and performing diffusion in that space. In standard Latent Diffusion Model (LDM), the latent code is formed by sampling from a Gaussian distribution (i.e., combining both the mean and the standard deviation), which helps regularize the latent space but appears to contribute little beyond the deterministic component. Motivated by recent empirical observations that the decoder relies primarily on the latent mean, our work reexamines this paradigm and proposes a decoupled latent diffusion model that focuses on a simplified latent representation. Specifically, we compare three configurations: (i) the standard latent code, (ii) a concatenated representation that explicitly preserves both mean and variance, and (iii) a deterministic mean-only representation. Our extensive experiments on multiple benchmark datasets demonstrate that, when compared to the standard approach, the mean-only configuration not only maintains but in many cases improves synthesis quality by producing sharper and more coherent images while reducing unnecessary noise. These findings suggest that a simplified, deterministic latent representation can yield more stable and efficient generative models, challenging the conventional reliance on latent sampling in diffusion-based image synthesis.https://ieeexplore.ieee.org/document/11091282/Denoising diffusion modellatent representationimage generation
spellingShingle Hyun-Tae Choi
Kensuke Nakamura
Byung-Woo Hong
Decoupled Latent Diffusion Model for Enhancing Image Generation
IEEE Access
Denoising diffusion model
latent representation
image generation
title Decoupled Latent Diffusion Model for Enhancing Image Generation
title_full Decoupled Latent Diffusion Model for Enhancing Image Generation
title_fullStr Decoupled Latent Diffusion Model for Enhancing Image Generation
title_full_unstemmed Decoupled Latent Diffusion Model for Enhancing Image Generation
title_short Decoupled Latent Diffusion Model for Enhancing Image Generation
title_sort decoupled latent diffusion model for enhancing image generation
topic Denoising diffusion model
latent representation
image generation
url https://ieeexplore.ieee.org/document/11091282/
work_keys_str_mv AT hyuntaechoi decoupledlatentdiffusionmodelforenhancingimagegeneration
AT kensukenakamura decoupledlatentdiffusionmodelforenhancingimagegeneration
AT byungwoohong decoupledlatentdiffusionmodelforenhancingimagegeneration