AndyLone22

AndyLone22 / MirrorMetrics

Public

MirrorMetrics: How to evaluate Stable Diffusion LoRAs. A visual diagnostic tool to detect overfitting, check dataset quality, and fix training settings using InsightFace biometrics.

civitai.comarticles26241 ai-art computer-vision data-visualization dataset-analysis insightface
43
4
100% credibility
Found Feb 25, 2026 at 41 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

MirrorMetrics is a local analysis tool that compares reference face photos with AI-generated images to quantify identity consistency, pose stability, and other visual metrics via interactive dashboards and reports.

How It Works

1
🔍 Discover MirrorMetrics

While playing with AI tools to generate consistent faces from photos, you stumble upon this helpful analyzer that measures how well the AI captures the real person's look.

2
💻 Get it ready on your computer

Follow a few easy steps to set up the tool so it works smoothly on your machine, like creating a safe space for it to run.

3
📁 Sort your photos into folders

Place your real-life photos in the reference folder and group your AI-generated images into separate folders for each test version.

4
🚀 Start the magic

Double-click the starter file or run it simply, and watch as it scans all your images to uncover hidden patterns.

5
📊 View the colorful dashboard

Open the new web page in your browser to see interactive charts full of graphs and data about face matches and poses.

6
👀 Spot the insights

Examine similarity scores, age estimates, angle variety, and copycat matches to understand strengths and weaknesses.

✅ Master your AI creations

Celebrate knowing exactly which AI versions nail the identity, detect overtraining, and guide your next improvements.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 41 to 43 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is MirrorMetrics?

MirrorMetrics is a Python tool that lets you evaluate Stable Diffusion LoRAs using InsightFace biometrics for face identity consistency. Drop your training dataset into a Reference_Images folder and generated images from different LoRAs into subfolders under Lora_Candidates, then run a simple script to get an interactive Plotly dashboard plus a Copycat report. It detects overfitting, checks dataset quality, and reveals issues like poor pose variety or dominant training samples to help fix training settings.

Why is it gaining traction?

It stands out with local-only processing for privacy, no cloud uploads, and rich diagnostics like leave-one-out similarity scores, t-SNE identity maps, and a visual "black hole" ranking of overused dataset images. The Copycat Detector pairs each generation with its nearest reference to spot memorization instantly, while pose scatter plots quantify flexibility across angles—features that turn vague training hunches into hard data. Developers hook on the zero-setup venv install and standalone CUDA support, even on CPU.

Who should use this?

Stable Diffusion artists training face or character LoRAs who waste epochs on bad datasets or rigid outputs. Dataset curators checking for outliers before training, or trainers comparing LoRA variants from different steps, epochs, or prompts to pick winners. Ideal for anyone generating diverse poses and needing biometrics to validate diffusion model generalization.

Verdict

Grab it if you're deep in LoRA workflows—solid docs and outputs make it immediately useful despite 41 stars and 1.0% credibility signaling early maturity. Test on your next run; lacks broad adoption but delivers targeted value without fluff.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.