---
title: MemGuard
emoji: 🛡️
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 6.19.0
python_version: '3.11'
app_file: app/demo.py
pinned: false
license: mit
short_description: Memorization detection for Stable Diffusion v1.4
---

# 🛡️ MemGuard

Detect **training-data memorization** in Stable Diffusion v1.4 outputs.

Enter a prompt → MemGuard generates the image and scores it with the
anisotropy-of-log-probability metric from
[Detecting and Mitigating Memorization in Diffusion Models through Anisotropy of the
Log-Probability](https://openreview.net/forum?id=HTPGy5ydAY) (ICLR 2026). Scores
at/above the **0.9** threshold are flagged as memorized.

**Source code, library, API & Docker:** https://github.com/rohanasthana/memguard

> **Note:** this Space runs on free CPU hardware, so image generation is slow
> (~100 s/denoising step). Lower the image-steps slider to finish sooner.