AZET Blog

AI for Annotating Tidepool Sketches

October 18, 2026

As of 2026-10-16 · AZET 운영팀 (the AZET team). We operate azet.io and publish this guide for fun and reference.

Four arrows point into a tidepool sketch. One marks a pale edge, one marks a dark patch, one points to broken lines used to show uncertainty, and the fourth has no explanation. We think the useful editing job starts with those arrows, before anyone tries to make the page sound like a nature guide.

Every sketch and sample annotation in this article is fictional. This is a notebook exercise, not an identification guide. If you use an AI text assistant, give it your own descriptions of the drawing and what you observed. Do not assume it has inspected the scene or your page.

Build a key before writing captions

For the fictional sketch, imagine a page containing an outlined pool, a rounded shape near the left edge, a dark patch near the middle, and several broken lines on the right. The observer has written these notes separately:

  • A points to the rounded shape: pale outline; exact edge difficult to follow.
  • B points to the dark patch: looked dark while sketching; texture not recorded.
  • C points to the broken lines: these marks indicate uncertainty, not a pattern I observed.
  • D is an unlabeled arrow: I cannot now remember what I meant.

A stable letter key keeps a caption connected to a mark. Use letters already on the page, or add them yourself. The assistant should not invent where an arrow points from a loose description.

There are also two sources here: the marks on the paper and the observer's notes about the scene. They may differ. A drawn boundary can be sharp even when the observer remembers a blurred edge. An annotation should preserve that distinction.

Draft short labels in two passes

First, ask for plain labels that fit beside the drawing. Then ask for a longer margin key where uncertainty has room to breathe. Trying to fit everything into three words can strip out the most useful qualification.

A prompt for a general text assistant could be:

Using my lettered descriptions, propose a short label and a longer margin note for each letter. Describe only the features I supplied. Keep drawing conventions separate from scene observations. Do not identify organisms, infer materials, or fill an unexplained arrow. Mark missing information explicitly. Treat all examples as hypothetical.

For our fictional page, a suitable editorial target would look like this:

Key Short label Margin note
A Unclear pale edge Rounded shape; exact boundary was difficult to follow
B Dark-looking patch Darkness recorded; texture was not recorded
C Uncertainty marks Broken lines are my drawing convention, not an observed pattern
D Meaning unresolved Arrow remains on page; intended subject not remembered

Those captions are deliberately modest. They make the drawing usable without converting every shape into a named object. The longer note also explains why the marks look the way they do.

Catch the caption that changes the drawing

Suppose a hypothetical draft calls C "striped surface." It sounds concise, but it changes a graphic convention into an observation. The person reviewing the page should replace it with "uncertainty marks" and keep the explanation in the margin.

A second hypothetical draft calls B "rough dark rock." The source notes supply neither roughness nor material. Remove both additions. "Dark-looking patch" is sufficient until the observer supplies more information.

Check each noun as carefully as each adjective. "Shell," "plant," or "animal" can introduce an identification even when the rest of the caption sounds cautious. If the original record says only "rounded shape," a more specific noun needs a separate basis supplied by the person.

Make the page work at two reading distances

Read the short labels alone. They should help you locate the corresponding marks. Then read the margin notes without looking at the sketch. They should explain the uncertainty and the limits of the description.

If two labels could point to the same feature, add a location from your own drawing description, such as "left edge." Avoid adding measurements or scale unless you recorded them. This fictional page has no scale, so its caption key makes no size comparison.

Keep any later correction visibly separate. You might append "later clarification" with your own explanation instead of replacing the old label. The page can then show both the original drawing decision and the later interpretation without blending them.

FAQ

What should I do with an arrow I cannot explain?

Keep it visible and mark its meaning unresolved. Erasing it may remove a useful reminder; inventing a caption adds a detail you never recorded. If you later remember its purpose, attach that recollection as a later note rather than presenting it as original evidence.

Can I ask for scientific names for the shapes?

That is outside this annotation exercise. A sketch description does not establish an identification. Keep your descriptive key intact, and place any separately researched candidate names in a distinct section with their source and uncertainty, if you decide to investigate further.

How short should a label be?

Short enough to fit comfortably beside its arrow, but long enough to retain the important qualifier. Put the explanation in the margin key. In the fictional example, "unclear pale edge" carries more of the observer's meaning than "pale edge," even though it takes another word.

Our product reference

The 2026-10-03 maker introduction describes azet as an AI assistant and describes an email early-access waitlist. This article offers a general text-editing exercise; it does not document an AZET sketch-analysis feature.

If that dated waitlist introduction interests you, visit https://azet.io for the waitlist and check the website for the latest availability.

tidepool sketchesannotationsdrawingobservationAI writing