We just finished building an internal prompt library for our design team to use with Midjourney. It's mainly to keep our brand visuals consistent across different projects.
The basic structure is a simple Notion database. We have fields for the project name, the specific style we're aiming for (like "product shot" or "iconography"), the full prompt text, and the seed number. We also link to the final approved image. It's super basic, but it's already saving us from constantly digging through old Discord channels.
I'm curious how others handle this. Do you use a different tool? Are there any other fields we should be tracking?
Notion's a solid start for a simple library, especially with its linking capabilities. You're right to track the seed. I'd strongly suggest adding a field for the Midjourney model version (like --v 5.2 or --niji 6). Output characteristics can drift significantly between versions, and that's crucial for reproducibility.
Consider adding metadata fields for generation parameters beyond the seed. Things like aspect ratio (--ar), stylization (--s), and chaos (--c). Without these, you're only capturing part of the recipe. A simple text field for "full command" that includes all the switches might be more future-proof than trying to parse them individually.
From an operational standpoint, have you considered how you'll handle versioning? If a prompt is iterated on ten times, does each iteration become a new database entry? You might want a "parent prompt" link field to track lineage, or a simple version number. It prevents fragmentation of similar attempts.
—chris