mysnipbox
← All articles

Tidy Up Your Prompt Library: Seiri, Seiton and Shūnō for AI Image Creators

Published November 2, 2026 · 6 min read

Japanese has everyday words for tidying that English lacks a neat match for. Seiri (整理) means sorting what you need from what you don't. Seiton (整頓) means giving every item one fixed place. Shūnō (収納) means storing things so they are easy to take out again. The factory-floor "5S" method starts with the first two.

An AI image library needs exactly this. After a few months you have thousands of generations, half of them near-duplicates, and a prompt you can't remember why you saved. Here is a three-step tidy-up that follows the way mysnipbox is organized.

1. Seiri (整理): decide what stays

Don't try to organize everything. First decide what is worth keeping, using three questions for each generation:

  1. 1Would I reuse the look? A lighting setup, a style or a composition you would want again counts. A one-off subject usually doesn't.
  2. 2Is it hard to get back? If the prompt would take ten minutes to reconstruct, keep it. If it is only quality-tag boilerplate, it isn't worth a spot.
  3. 3Is it the clearest take? Of five near-identical results, keep the one that shows the look most clearly.

In mysnipbox, new images land in the Inbox, a temporary holding area where you review them before filing them away. Treat it as your sorting table: nothing moves on to All Vault until it passes the questions above.

2. Seiton (整頓): give every piece a fixed place

Tidying works because every item has one home. For prompts, those homes are the six categories every tag is sorted into. Think of them as six drawers:

Drawer (category)What lives thereTidy-up rule
Subjectcybernetic samurai, vintage roadsterUsually one-off, so don't over-collect
Style & Medium3d digital render, claymation textureYour signature look: keep the best
Action & Posefront view, centered compositionReuse freely
Attire & Detailsblack turtleneck sweater, knitted blue scarfKeep what belongs to a recurring character
Environmentseamless studio backdropKeep a handful of favorite backdrops
Lighting & Palettesoft studio diffusion, vibrant pastel paletteThe most reusable drawer

Then separate the cabinets. Personas keep one character, client or visual style apart from the rest, so the Vault, the tag filter and Tag Insights only show that persona's material. One persona per project is a good start, and you can add more later. See Personas & filtering.

3. Shūnō (収納): store it so you can see it

Good storage means finding things without opening every box. Instead of searching by file name, search by what is in the image:

  • Filter by tags: the tag bar above the gallery lists every tag in your Vault, grouped by category, with how many images use it. Click a tag to narrow down, click more to narrow further.
  • Check the look in Tag Insights: see thumbnails of what a tag actually looks like in your own images, and which tags are used together.
  • Re-decompose thin results: if an image's tags look sparse, run Re-decompose with Vision LLM with a stronger model.
  • Turn favorites into templates: click your best tags into the Prompt Basket, add a {subject} variable and save it as a template.
Tag Insights shows which tags travel together and what a tag looks like in your own images (thumbnails blurred here).

4. Keep it tidy: a 5-minute weekly routine

  1. 1Open the Inbox and ask the three seiri questions of each new image.
  2. 2File the keepers under the right persona and check that their tags look right in the six categories.
  3. 3Spot a lighting or style tag that keeps coming back? Open Tag Insights and turn it into a template.
  4. 4Leave the Inbox empty.

Everything stays in a local SQLite database on your computer, so tidying doesn't mean uploading your library anywhere. Cloud providers only receive the images you choose to analyze; for fully on-device processing use local models. More on the Vault in Organize your Vault.

Want to feel the flow first? The browser demo runs the tag filter and the Prompt Basket with sample images.