Vexley Versioning System

A lightweight naming and organization convention to preserve facial fidelity and character essence in iterative AI generation workflows.

The Vexley Versioning System provides a traceable, drift-aware file naming and folder convention for creators who maintain consistent AI persons, companions, virtual idols, or narrative characters from a single foundational reference image. It combats cumulative identity degradation across img2img, video-frame extraction, upscaling, and chained generations.

About

X: @5ynthaire
GitHub: https://github.com/5ynthaire
Mission: Transcending creative limits through human-AI synergy
Attribution: Developed with Grok 4.1 by xAI (no affiliation).
Origin of name: Named after Eurybia Harlow-Vexley, the first AI character this versioning system was created to protect.

Core Concepts

Reference Image

Also known as: Canonical Reference, Genesis Image
The single original image that defines the character’s core identity—facial structure, vibe, and essence. All derivatives trace back to this file or its descendants.
Preferred term: “Reference Image” (tool-agnostic and widely used in --cref, IPAdapter, PuLID, FaceID contexts).

Drift Degree

Also known as: Iteration Degree, Separation Degree
The generational distance from the reference image.

• 00 = the reference itself (zero drift)
• 01 = direct outputs from the reference
• 02 = outputs from a 01 file, etc.
  Higher degrees increase risk of fidelity loss.

Derivative

Also known as: Offspring, Iteration
Any generated asset descended from the reference or a prior derivative.

Drift Minimization Practices

• Always use the 00_Reference (or strongest low-degree file) as the primary --cref / IPAdapter / PuLID input.
• Apply low denoising strengths (typically 0.25–0.45) for controlled evolution.
• Prioritize consistency mechanisms: IPAdapter-FaceID, PuLID-Flux, InstantID, ControlNet reference-only.
• Train a lightweight LoRA on the reference image early to create a stable anchor.

Vexley Versioning System Version 1

Filename Format

[drift-degree]_[YYYYMMDD]_[Platform]-[SessionID]_[optional-descriptor].[ext]

• [drift-degree] — Exactly two zero-padded digits (00–99). Sorts lowest-drift files first.
• YYYYMMDD — Creation/download date.
• Platform — Tool/service (GrokImagine, Midjourney, Flux, ComfyUI, Perchance, etc.).
• SessionID — Short reproducibility token:
  • Web: last portion of generation URL/hash
  • Local (SD/Comfy): truncated seed (last 8 chars typical)
  • Other: short UUID or timestamp fragment
• optional-descriptor — Free-text, prefixed by underscore or hyphen.
  Recommended: -parent[prev-degree]-[prev-sessionID] for explicit lineage.
  Examples: _smiling, _cyberpunk-loft, _defiant-pose.

Examples

• 00_20251213_Perchance-Original.jpg
• 01_20251227_GrokImagine-abc123.jpg
• 01_20251228_GrokImagine-def456-parent00.jpg
• 02_20251228_GrokImagine-ghi789-parent01-abc123.jpg
• 02_20251229_GrokImagine-jkl012-parent01-def456_smiling-loft.mp4

Folder Structure

Group files by drift degree for natural alphabetical sorting and tree readability:

/CharacterName/
  00_Reference/
    00_20251213_Perchance-Original.jpg
  01_Direct/
    01_20251227_GrokImagine-abc123.mp4
    01_20251228_GrokImagine-def456-parent00.jpg
  02_SecondLevel/
    02_20251228_GrokImagine-ghi789-parent01-abc123.jpg
    02_20251229_GrokImagine-jkl012-parent01-def456_smiling.mp4
  03_ThirdLevel/
    …

Lineage Tracking Methods

1. Filename-embedded (simplest)
   Use -parentXX-YYYY in the descriptor field.

2. Sidecar metadata files (recommended for complex trees)
   Small YAML/JSON beside each file or per folder:

yaml
name: 02_20251228_GrokImagine-ghi789.jpg
drift_degree: 02
parent: 01_20251227_GrokImagine-abc123.jpg
platform: GrokImagine
session_id: ghi789
prompt: "Eurybia Harlow-Vexley in tech noir loft, smoking, defiant pose..."
notes: "Highest facial fidelity retained"

3. Embedded file metadata (most durable)
   • Local: Save PNGs with full generation data (native in ComfyUI/SD).
   • Web: Inject via ExifTool (prompt, parent filename, notes, etc.).

Future Updates and Versioning Rationale

Major platform changes (native lineage metadata, standardized generation provenance fields, new consistency tools adopted at scale) may justify schema updates.

When warranted, a new version (Version 2, etc.) will be released with:

• explicit changelog
• migration guidance
• backward compatibility where practical

Changes will remain infrequent and driven only by ecosystem-level improvements that meaningfully impact long-term traceability or fidelity preservation.

License

Released under the MIT License. Free to use, modify, and distribute.