Collecting What Works

Every professional image generator I know maintains a reference library. Not a folder of bookmarks — a structured collection of successful prompts paired with their outputs, reference images that inspired them, and notes on what made each one work. The library is the compound interest of image generation. Every successful prompt becomes a template for the next project. Every failed experiment teaches you what a specific model cannot do.

1. Save the Prompt and the Output Together Never save an image without the prompt that created it. Use the image filename or a sidecar text file. If you cannot reproduce the image, you cannot learn from it.
2. Collect Reference Images from the Real World When you see a photograph, illustration, or design that captures a mood, lighting setup, or composition you want to reproduce — save it. Screenshots from films, pages from design books, photographs from magazines. These become your visual vocabulary.
3. Tag by Visual Element, Not by Subject Organize by what makes the image work: "dramatic side lighting," "shallow depth of field," "muted earth tones," "overhead flat lay composition." Subject-based organization ("dogs," "landscapes") is useless when you need a specific visual technique.
4. Note What Failed Keep a "failure log" of prompts that consistently produce bad results. Knowing that a specific phrasing confuses a specific model saves you from repeating the experiment.

Start with a simple folder structure: /prompts/ for successful prompt-output pairs organized by style, /references/ for collected real-world inspiration images, and /failures/ for documented dead ends. The structure does not matter as much as the habit. Save everything. Annotate everything. Future you will thank present you when a client asks for "that warm film look we used three months ago" and you can pull the exact prompt in ten seconds.