Three Kinds of Notes to Keep Beside an AI Character Draft

A short note beside a character image can be more helpful than a confident label such as finished or approved. The label hides the question that was answered, while the note can explain what someone actually saw and what they still need to decide. This matters when AI-assisted images move from a private experiment into a shared creative project. The image may be useful, but its appearance alone does not tell another person which interpretation, revision or intended use belongs with it.

Three questions that should not share one answer

Consider a fictional shopkeeper intended for a small village scene. One person wants the character to look welcoming, another checks whether a bag is distinguishable from the coat, and a third needs a file to place in a prototype. These people are doing related work, but they are not answering the same question. A single status mark cannot reliably express all three judgments. Separating observations, intentions and open questions makes the handoff clearer without requiring a complicated scoring system.

An observation describes something that was examined under stated conditions. An intention describes what the project wants the image to communicate or achieve. An open question names work that has not yet been completed. These categories are deliberately modest. They do not certify quality, determine ownership or replace an integration test. Their purpose is to help a later reader understand the relationship between a particular image and a particular piece of reasoning, rather than infer a broad promise from an attractive preview.

Keep the tool description factual

Sprite Maker is a browser-based tool for AI-assisted character images, direction views and NPC sets, with local pixel editing and transparent PNG export. A sprite creator can supply a candidate for the exercise described here, but the notes remain a separate human documentation practice. Local editing and export are free. AI generation requires sign-in and credits and may involve introductory usage or paid options. That distinction matters because access to an editor is not the same as unlimited access to generation.

Neither a product description nor a downloaded PNG proves that a character is ready for every game project. Animation continuity, project conventions, import behavior and commercial-use rights require their own consideration. This article does not claim that the application implements the three-note framework or automatically validates those matters. The framework is an educational way to describe a creative handoff. It can be used with a manually selected export and ordinary written notes without inventing a software feature or implying that an unperformed test has already passed.

Write observations that have a visible subject

Start an observation by identifying its subject. A filename and revision are usually more useful than a loose phrase such as the latest character. Then state the narrow thing that was inspected and the conditions that affected the view. For example, a hypothetical note might concern whether the shopkeeper's bag can be distinguished from the coat against one light background at a chosen display size. The example is a model of wording, not a report that this review has been performed on an image shown here.

Avoid turning the observation into a diagnosis that the evidence cannot support. A pale edge that is difficult to see on one sample background does not establish that generation is broken. It identifies a possible readability issue under those conditions. The next person can inspect the edge, try a small adjustment or compare a different candidate. If the file changes, attach subsequent observations to the new revision. Keeping the original export makes the sequence understandable and avoids silently transferring an old judgment to an image that was never examined.

Describe intentions without presenting them as results

An intention is useful even when it has not yet been achieved. The fictional shopkeeper might be meant to appear approachable, remain distinct from nearby scenery or carry an accessory that explains a role. Those goals guide later choices, but they are not evidence that the candidate already satisfies them. Write the intention as a desired outcome and leave space for uncertainty. A sentence that begins with the project needs is often less misleading than one that declares the image is perfect for the project.

Intentions also help resolve disagreements that appear to be about taste. One reviewer may favor more detail while another wants a simpler silhouette. If the intended viewing size and scene role are not stated, both suggestions can sound reasonable while pulling the image in opposite directions. Naming the purpose does not force agreement or prove one person's opinion correct. It makes the disagreement easier to discuss and creates a clearer basis for the next bounded change, rather than encouraging a sequence of unrelated edits.

Make open questions useful enough to act on

An open question should name a decision or check that someone can actually undertake. Confirm whether the chosen directions preserve an important accessory is more actionable than check everything. Likewise, determine how the project expects transparent exports to be named is more precise than make it compatible. Keep the question proportional to the handoff. A small draft need not pretend to have a complete production checklist, but it should not hide important unknowns behind a confident-looking label or an unexplained green icon.

Distinguish visual questions from later technical work. Looking at several still images can reveal visible differences, but it does not establish frame timing or behavior in a game engine. Downloading a PNG does not prove that a particular importer accepted it. If a later integration check is performed, record the actual procedure and result separately. Until then, leave the question open. This is not an admission that the candidate is unusable; it is a more accurate description of what the current evidence does and does not cover.

Keep the three notes attached to a revision

A practical handoff can fit in a small text document beside the export. Give it a clear title, identify the image, and use three short sections for observation, intention and open questions. Include the viewing conditions where they matter. Do not use the note as a substitute for retaining the source or the exported file. Its value comes from explaining a specific artifact. If the file is renamed or replaced, update the relationship explicitly rather than hoping a later reader will reconstruct it from a folder of similar images.

The accompanying illustration should be identified just as carefully. A project graphic can explain which tool is being discussed, but it is not automatically the character that received a review. Do not treat a header image as evidence of a completed inspection. Where a real example is shared later, name the actual candidate and state which parts of the note concern it. This keeps explanatory material separate from results and prevents a convenient visual from carrying claims that the written text was careful not to make.

End with a decision rather than an invented score

The next step might be another pixel edit, a different candidate, a clearer statement of the scene role or a separate technical check. The three-note approach does not choose that step automatically. It gives the people involved enough context to choose it honestly. A limited observation can remain useful even when the intention changes, and an open question can guide new work without being mistaken for a failure report. The result is a more readable creative handoff, not a guarantee that an asset is production-ready.

This article was prepared with AI assistance in connection with Sprite Maker. It is not an independent review, a report of measured performance or a claim that integration tests were performed. The shopkeeper scenario and sample notes are illustrative. No certification, ownership determination or commercial-use clearance is asserted. The recommendation is simply to preserve the candidate, distinguish evidence from desired outcomes, and give unresolved work a clear name so that another person can continue from the actual state of the project.