BodyRec: Open‑Source Body Fingerprinting Tool
Run BodyRec on a folder of full‑body renders to generate fingerprints and rank your library offline.
Run BodyRec on a folder of full‑body renders to generate fingerprints and rank your library offline.
Summary
BodyRec is an open‑source Python tool released by user FugueSegue that fingerprints full‑body renders to enable offline similarity ranking. The GitHub repository (https://github.com/FugueSegue/bodyrec) stores a “body fingerprint” consisting of relative bone lengths via NLF and SMPL shape coefficients via GVHMR, kept as two separate similarity scores – skeleton and build – rather than a single blended metric. The tool requires a ComfyUI server with the NLF and GVHMR node packs and runs on Windows with Python 3.11; it ingests a folder of full‑body renders, generates about twelve renders per character to compute a median fingerprint, and outputs per‑segment breakdowns. BodyRec can rank an entire library offline, providing a clean similarity score that is robust to seed variation and pose framing. The design choices were driven by two findings: LoRA renders vary with seed, so a median over ~12 renders is used, and pose‑based shape estimation is unreliable, so clean full‑body renders are required. The project is MIT‑licensed, with underlying models licensed for research/non‑commercial use, and is intended as a discussion starter rather than a production product.
Key changes
- Stores body fingerprint with relative bone lengths via NLF and SMPL shape coefficients via GVHMR
- Keeps skeleton and build as separate similarity scores, not blended
- Computes median fingerprint over ~12 renders per character to mitigate seed variation
- Requires ComfyUI server with NLF and GVHMR node packs, runs on Windows/Python 3.11
- Provides offline ranking of entire library with per‑segment breakdowns
- MIT‑licensed, underlying models research/non‑commercial