The image curation methodology for AI model training

The problem with your AI images is the source.

For creative professionals using gen-AI who want outputs that actually look like them.

PROBLEM STATEMENT

Most AI image tools disappoint people.

The outputs look plastic, generic, or just wrong — like an approximation of you that was assembled from a description rather than built from who you actually are.

The instinct is to fix the prompts. Write more detail. Try different language. Generate again.

The prompts are not the problem.

AI model training is pattern recognition. Feed it inconsistent, low-quality, or stylistically incoherent images and it will faithfully learn the wrong things — and produce the wrong things back. The tool is doing exactly what it was trained to do. The source images are the issue.

WHAT IT IS

Source Curator is the curation methodology for training AI models that actually look like you.

It covers everything that happens before you generate — how to define your aesthetic, how to build a source image set with the right mix of settings, framing, angles, and lighting conditions, and how to identify and fix drift when your model stops producing accurate outputs.

The methodology works for Adobe Firefly. The principles apply to any model training platform.

HOW IT WORKS

1. Define your aesthetic
Before you touch a single image, decide what you want your outputs to feel like. That decision becomes your filter for every image you keep and every image you cut.

2. Build your source set
Thirty images, structured with genuine variety across settings, framing, pose, angle, and lighting. The curation framework tells you exactly what the mix needs to look like — and what to cut without hesitation.

3. Train, test, refine
Upload your curated set, run your model against the master prompt, and evaluate honestly. If something is off, the Drift Diagnostic tells you why and how to fix it.

  • Bonus 01 — The Prompt Library

    One master test prompt and 30 generation prompts across varied poses, angles, lenses, and lighting conditions. Written and tested against a working personal model. Use them as-is or adapt them to your own subject and style.

  • Bonus 02 — The Curation Checklist

    Every selection criterion from the methodology in one scannable page. Run through it before you upload your image set.

  • Bonus 03 — The Drift Diagnostic


    A troubleshooting guide for when your model isn't performing as expected. Symptoms, likely causes, and the specific fixes — so you're not guessing when something looks wrong.

  • This is for you if:

    — You're using Adobe Firefly's personal model training and outputs don't look like you
    — You've tried adjusting your prompts and it hasn't solved the problem
    — You want to understand the curation logic before you train or retrain a model
    — You're a creative professional who produces content and needs consistent, on-brand imagery

  • This is not for you if:

    — You haven't started using AI image generation yet — start there first
    — You're looking for a prompt library without the methodology behind it
    — You want outputs in a style that doesn't reflect your actual aesthetic — this is about accuracy, not transformation

ABOUT CATHERINE

Catherine Gomersall is a creative professional with 25 years of experience across design, systems, and IP commercialisation. She is the founder of Quartz Ripple and the creator of the Creative Integration Method — a framework for helping experienced creative professionals turn what they know into income.

Source Curator came directly from her own practice. The Eclectic Lifestyle Photography model visible across the Quartz Ripple brand was built using this methodology.

View my Model Portfolio