Two years ago a designer's saved material was screenshots and links. Now a large share of it is prompts: the exact words that produced an illustration, a texture, a hero image or a dozen layout ideas. And prompts are being lost faster than screenshots ever were.
They live in Discord scrollback, in a Notes app, in the history panel of whichever tool made them. The output lands in Downloads. The reference that inspired the prompt is in a third place. Three weeks later you have the image, cannot find the prompt, and start again from scratch.
The rule: prompt, output and source stay together
A prompt without its output is a sentence. An output without its prompt is unrepeatable. A pair without the reference that inspired it has lost its lineage. All three belong in one record, in the same library as everything else you save, so one search finds the set.
In practice that means your reference manager needs a prompt type, not just images and links. GetRef treats prompts as first-class saves, so a prompt sits next to the images it produced and the screenshot that started the idea, and all of it is searched together.
What to record with every prompt
- The full prompt text, including parameters and negative prompts.
- The tool and model version. Results differ between versions; a prompt that worked in one may not in the next.
- The output you kept, as an image, ideally the chosen variation rather than the whole grid.
- One line on what worked. "Lighting is right, hands are not." That sentence is more useful than the prompt itself a month later.
Tagging prompts
The four-axis tagging system works for prompts with one change: the what-it-is axis becomes subject and style. isometric, illustration, texture, portrait, icon-set on one side; flat, painterly, 3d, line-art, photographic on the other. Add the tool as a tag too. Most of this can be read automatically from the prompt text itself.
Searching prompts
Search by what the output looks like, not by the words in the prompt. You will remember "the warm isometric office scene", not the exact parameter string. A tool that indexes the prompt and the image together handles both, and AI image search covers the cases where you only remember the look.
A simple routine
- When a generation is worth keeping, save the image and paste the prompt into the same record.
- Tag tool, subject and style. Let automatic tagging handle colour and content.
- Write the one line on what worked.
- Once a month, delete the experiments that went nowhere.
Everything else about keeping a library applies unchanged: one home, one click to save, no manual filing. See how to build a design library for the framework the prompts sit inside.
Common questions
- Where should designers store AI prompts?
- In the same library as the images they produced and the references that inspired them. A prompt on its own, in a notes app, loses the one thing that made it valuable: the result. Keep the prompt, the output and the source reference together.
- How do I organize Midjourney prompts?
- Save the prompt as text alongside the generated image, tag it by subject, style and model version, and add one line about what worked. Then search by description later rather than scrolling a Discord channel.
- Can GetRef save prompts?
- Yes. Prompt is a save type alongside links, images and notes, and prompts are tagged by their content so a search such as "isometric fintech illustration" finds the prompt that produced it.
