Interpreting the Weight Space of Customized Diffusion Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Dravid, Amil, Gandelsman, Yossi, Wang, Kuan-Chieh, Abdal, Rameen, Wetzstein, Gordon, Efros, Alexei A., Aberman, Kfir
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915030408101888
author Dravid, Amil
Gandelsman, Yossi
Wang, Kuan-Chieh
Abdal, Rameen
Wetzstein, Gordon
Efros, Alexei A.
Aberman, Kfir
author_facet Dravid, Amil
Gandelsman, Yossi
Wang, Kuan-Chieh
Abdal, Rameen
Wetzstein, Gordon
Efros, Alexei A.
Aberman, Kfir
contents We investigate the space of weights spanned by a large collection of customized diffusion models. We populate this space by creating a dataset of over 60,000 models, each of which is a base model fine-tuned to insert a different person's visual identity. We model the underlying manifold of these weights as a subspace, which we term weights2weights. We demonstrate three immediate applications of this space that result in new diffusion models -- sampling, editing, and inversion. First, sampling a set of weights from this space results in a new model encoding a novel identity. Next, we find linear directions in this space corresponding to semantic edits of the identity (e.g., adding a beard), resulting in a new model with the original identity edited. Finally, we show that inverting a single image into this space encodes a realistic identity into a model, even if the input image is out of distribution (e.g., a painting). We further find that these linear properties of the diffusion model weight space extend to other visual concepts. Our results indicate that the weight space of fine-tuned diffusion models can behave as an interpretable meta-latent space producing new models.
format Preprint
id arxiv_https___arxiv_org_abs_2406_09413
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Interpreting the Weight Space of Customized Diffusion Models
Dravid, Amil
Gandelsman, Yossi
Wang, Kuan-Chieh
Abdal, Rameen
Wetzstein, Gordon
Efros, Alexei A.
Aberman, Kfir
Computer Vision and Pattern Recognition
Graphics
Machine Learning
We investigate the space of weights spanned by a large collection of customized diffusion models. We populate this space by creating a dataset of over 60,000 models, each of which is a base model fine-tuned to insert a different person's visual identity. We model the underlying manifold of these weights as a subspace, which we term weights2weights. We demonstrate three immediate applications of this space that result in new diffusion models -- sampling, editing, and inversion. First, sampling a set of weights from this space results in a new model encoding a novel identity. Next, we find linear directions in this space corresponding to semantic edits of the identity (e.g., adding a beard), resulting in a new model with the original identity edited. Finally, we show that inverting a single image into this space encodes a realistic identity into a model, even if the input image is out of distribution (e.g., a painting). We further find that these linear properties of the diffusion model weight space extend to other visual concepts. Our results indicate that the weight space of fine-tuned diffusion models can behave as an interpretable meta-latent space producing new models.
title Interpreting the Weight Space of Customized Diffusion Models
topic Computer Vision and Pattern Recognition
Graphics
Machine Learning
url https://arxiv.org/abs/2406.09413