Table of Links
Abstract and 1. Introduction
-
Background and Related Work
-
Threat Model
-
Robust Style Mimicry
-
Experimental Setup
-
Results
6.1 Main Findings: All Protections are Easily Circumvented
6.2 Analysis
-
Discussion and Broader Impact, Acknowledgements, and References
A. Detailed Art Examples
B. Robust Mimicry Generations
C. Detailed Results
D. Differences with Glaze Finetuning
E. Findings on Glaze 2.0
F. Findings on Mist v2
G. Methods for Style Mimicry
H. Existing Style Mimicry Protections
I. Robust Mimicry Methods
J. Experimental Setup
K. User Study
L. Compute Resources
J Experimental Setup
This section describes our general experimental setup and specifies the settings and hyperparameters of the methods we use. When possible, we use default values from the machine learning literature. For implementation details see our official repository: https://github.com/ethz-spylab/ robust-style-mimicry
J.1 Style Mimicry Experimental Details
As described in Section 3, our threat model considers style mimicry with a latent diffusion text-toimage model M that is finetuned on a set of images X in a style S. This section specifies our choices for model M, images X, style S, the hyperparameters for finetuning M, and the hyperparameters for generating images with the finetuned model. Where possible, we try to replicate the style mimicry setup used by Shan et al. to evaluate Glaze, and highlight any differences.
Model We use Stable Diffusion version 2.1 (Stability AI, 2022), the same model used to optimize the protections we evaluate (Shan et al., 2023a; Liang et al., 2023; Van Le et al., 2023).
J.2 Protections Experimental Details
Next, we describe specific hyperparameters we use to reproduce each of the protections.
J.2.1 Anti-DreamBooth
J.2.2 Mistϕ
J.2.3 Glaze
The Glaze authors were unable to share a codebase upon request. We thus use their publicly released Windows application binary. We use the latest available version of Glaze, v1.1.1. We set Intensity
to High and Render Quality to Slowest, to obtain the strongest protections. Appendix E includes qualitative results on an updated version released after we concluded our user study.
J.3 Robust Mimicry Methods Experimental Details
J.3.1 Gaussian noising
We manually tune the Gaussian noising strength to σ2 = 0.05.
J.3.2 DiffPure
We use conditional DiffPure with the best-performing publicly available image generation diffusion model, Stable Diffusion XL 1.0 (SDXL) (Podell et al., 2023). We implement conditional DiffPure using the HuggingFace AutoPipelineForImage2Image pipeline. We use classifier-free guidance scale guidance scale = 7.5 with prompt P = Cx for image x. We manually tune the number of diffusion timesteps t via the strength pipeline argument to strength = 0.2.
J.3.3 IMPRESS++
[8] www.artstation.com
[9] @nulevoy is the first ArtStation artist that we experimented with. In our experiments, we found “nulevoy” a suitable choice for the special word w∗ and use it for all artists. We check that all of nulevoy’s images are published after the release date of LAION-5B to ensure that SD 2.1 has no prior knowledge about nulevoy’s style.
