Multimodal means using more than one kind of information, such as text, images or audio. This course makes an editable image description or transcript, the written words from a recording.

Start with an abstract diagram you created or your own voice reading a fictional script. A photo can reveal names; a recording can capture a private conversation. You do not need to identify people or guess sensitive traits.

A media register lists each sample's source, owner, allowed purpose and whether uploading it to a provider is permitted. Having a file does not automatically mean you may upload, publish or reuse it. If other people appear or speak, get appropriate permission. Licensing still matters for other material.

A retention policy says which copies are kept and when they are removed. Cover originals, temporary uploads, generated text, logs, backups and provider storage. Marking a record retired does not erase every copy. Explain the cleanup that actually exists.

First complete the introductory AI API course. Use Node.js 22 or newer; Node 24 is the current long-term support release. Required samples run locally without paid calls. Optional provider use needs your own credits. Keep public results free of filenames or unrelated personal details unless the task needs them.

Save the following as media-permission-register.mjs and run node media-permission-register.mjs.

const samples = [
  { id: 'diagram-1', kind: 'image', source: 'self-authored abstract diagram',
    purpose: 'learning description', hostedAllowed: true, status: 'active' },
  { id: 'private-recording', kind: 'audio', source: 'unreviewed recording',
    purpose: 'unknown', hostedAllowed: false, status: 'active' }
];
const eligible = samples.filter(s => s.hostedAllowed &&
  s.purpose === 'learning description' && s.status === 'active');
console.log(eligible.map(s => s.id));

Expected behavior: Only the reviewed abstract diagram is eligible for the stated hosted-processing purpose. This register decision does not itself upload a file.

1. Create a media register for one abstract image and one self-recorded fictional script.
2. Describe the processing purpose, expected output, permission basis and what must be removed before optional hosted use.
3. Write a retention table for original bytes, temporary copies, generated draft, corrected result and diagnostics.
4. Add a disallowed sample and test that the workflow stops before provider access.
5. Record any licensing or contributor permission in plain language.

Treat every locally readable file as eligible. Show the unreviewed recording entering the candidate list, then restore explicit purpose and processing-permission checks. Explain the difference between file access and allowed use.

Question: Does possession establish hosted-processing permission?

Answer: No.

Question: Must this project identify people?

Answer: No. Its task is description or transcription.

Question: Does a retired database record prove all copies were erased?

Answer: No. Retention and provider cleanup need separate evidence.

Deliver the sample register and retention plan. Keep the project limited to permitted fictional or self-authored materials and make the processing boundary visible before any optional API request.

Next lesson: Validate type, size and decode limits.

OpenAI data controls and W3C media accessibility. Reviewed 2026-10-07; provider retention must be checked for the selected endpoint and account.