AI for Planning Foley Sound Cues
As of 2026-10-16 · AZET 운영팀 (the AZET team). We operate azet.io and publish this guide for fun and reference.
A fictional character puts down a bag, searches it, and pulls out a folded map. "Add rustling" is an idea, but it is not yet a usable sound plan. Does the sound start when the bag opens? Does it continue under the dialogue? What tells the person making it to stop?
I would separate those decisions before choosing clever objects to produce the sound. If you use an AI text assistant, it can be asked to organize descriptions you supply. You remain the person who listens, watches, and decides whether an attempted sound belongs in the scene. All scenes, tests, and sample notes here are hypothetical.
Begin with the action you can point to
Imagine a short rehearsal clip called The Missing Map. For this fictional exercise, a person has written five visible events: a bag touches a table, a hand searches inside it, a folded map appears above the opening, the map unfolds, and the character holds it still. The supplied notes also place the first spoken word just after the map appears. These are the anchors for a cue plan.
Separate each anchor from the intended impression. "Hand enters bag" is an event. "A hurried search" is an interpretation. Both can be useful, but putting them in different fields prevents an impression from becoming an unclear start instruction.
Try a simple worksheet with five fields: visible event, desired impression, candidate action, start, and stop. Leave candidate action empty until you have something to test. A blank field is more useful than a confident description of a sound nobody has heard.
Use a prompt that preserves the missing pieces
Turn these supplied observations into a Foley planning worksheet. Keep visible events separate from desired impressions. Include start and stop triggers. List missing information as questions. Do not describe how an untested object sounds. Do not add timecodes, equipment, or actions that are absent from my notes. Observations: [paste notes].
You could then add your own candidate actions beside the worksheet. In this fictional example, the maker proposes gently handling a paper bag and moving a folded sheet. Those are options to try, not claims that either will match the picture.
The first written plan might be:
| Event | Candidate action | Start | Stop |
|---|---|---|---|
| Hand searches the bag | Handle the paper bag | Hand enters opening | Map becomes visible |
| Map unfolds | Move the folded sheet | First fold opens | Map is held still |
The bag landing is intentionally absent. The hypothetical maker has not selected a candidate for it. Label that event "unassigned" rather than pretending the table is complete.
Let the listening note change the plan
Suppose the maker supplies this fictional test observation: "The bag handling continues after the map appears and covers the first spoken word." That is specific enough to edit. A vague reaction such as "too messy" leaves several possible repairs open.
A proposed revision could end the search sound at the earlier moment when the hand closes around the map. But that earlier moment might not be visible in the scene. Human review must catch the mismatch between a neat instruction and an observable trigger.
The corrected note could instead say: "Finish the bag handling when the top edge of the map appears. Leave the following spoken word uncovered." The maker must test that instruction against the actual performance. Written timing is still a proposal until someone tries it.
For another hypothetical test, the folded sheet makes a sound the maker describes as "sharper than the movement looks." Keep that wording as a listening impression. Do not upgrade it to a technical diagnosis. Ask for two planning options: reduce the amount of handling, or leave the unfolding moment silent. Then compare both yourself.
Make an alternatives card
For each uncertain cue, write a small card containing the candidate, your observation, the proposed change, and the next comparison. For the fictional map, the card might read:
- Candidate: one folded sheet, opened once.
- Observation: the sound draws attention away from the line.
- Proposed change: try the scene without that cue.
- Comparison: watch both versions and note where attention goes.
This keeps a rejected idea useful without allowing it to drift back into the working plan. Ask an assistant to condense the card only after you have supplied the comparison result. It cannot substitute an imagined listening session for yours.
Keep selection and performance instructions separate. "Try silence" belongs on the alternatives card. Once the group chooses silence, the working cue sheet should clearly show that the unfolding has no assigned sound.
FAQ
Should every visible movement receive a sound?
Treat that as an editorial choice for the piece. In the fictional map scene, the group could leave the unfolding silent to keep attention on a line. Write down the reason for the choice so a later revision does not accidentally restore the discarded cue.
Can a text assistant choose the best household object?
You can request candidate objects to consider, but a written suggestion is not a listening result. Describe what you have available and what you hear during your own tests. Keep untried candidates labeled as proposals, and select only after comparing them in context.
What should I write when timing changes between takes?
Record the visible start and stop events first. If you are working with a fixed clip, add your own checked location references. For a changing performance, identify which action varies and agree on a cue people can follow rather than inventing a duration to hide the variation.
Where AZET fits in this guide
The AZET introduction dated 2026-10-03 describes azet as an AI assistant and an email early-access waitlist. It does not establish that AZET performed these sound-planning exercises. To find the waitlist described in that introduction, visit https://azet.io; check the website for the latest availability.