🎭 ELSAC

Eliz's LoRA Samples Analyzer & Curator — the free, open-source way to build a consistent character dataset for LoRA training.

ELSAC takes your raw character images, organizes them by similarity, scores them against your best references, and tells you exactly which ones are worth training on — so you spend less time staring at folders and more time creating.

✨ Why ELSAC?

🧠 From chaos to clusters.
CLIP + HDBSCAN group your images by visual similarity automatically.

🎯 Honest ranking.
Single, face, and multi-reference scoring sort your set into clear quality buckets.

🔬 Quality you can trust.
Blur, face-ratio, and head-yaw analysis export to clean CSV reports.

🤍 Free & private.
MIT licensed. Runs locally — your images never leave your machine.

🔬 The workflow

📁

Project Setup

Step 1

Name your character, drop in images, and resume anytime — progress is saved.

🔍

Clustering

Step 2

CLIP + HDBSCAN group similar images into clusters and outliers.

⭐

Single-Ref

Step 3

Rank everything against one image that best represents your character.

👤

Face Likeness

Step 4

Faces are detected, cropped & aligned before comparing — real facial consistency.

👥

Multi-Ref

Step 5

Combine 4–8 varied references into one averaged identity, then rank your set.

🔬

Analysis

Step 6

Blur, face-ratio & yaw metrics exported as CSVs for final filtering.

🖥️ Built for creators

Local & private — runs on your own machine.
GPU-accelerated — real CUDA support, including RTX 50-series (Blackwell). CPU works too.
Resume anytime — long curation sessions survive restarts.
Open & MIT-licensed — free forever, yours to read and modify.

🚀 Try it in minutes

Windows

setup.bat
venv\Scripts\activate.bat
python app.py

macOS / Linux

./setup.sh
source venv/bin/activate
python app.py

Then open http://127.0.0.1:7860 🎉

📚 Learn more

User Guide & Docs 🐙 GitHub