Start accessibility work before the final export: identify meaningful audio and visual information, review captions against the finished soundtrack, plan how essential visuals will be described, and test the files in the intended player. These are editorial production suggestions informed by W3C guidance, not announced Milan AI Festival submission rules.
Separate three different audience needs
A viewer may need access to speech and sound, a translation of dialogue, or an explanation of essential visual information. Those needs can overlap, but they are not interchangeable. W3C WAI’s captions guidance distinguishes same-language captions from translated subtitles and explains that captions also convey meaningful non-speech audio. Its visual-description guidance addresses information that people cannot obtain by listening to the original soundtrack alone. Both resources are linked below.
Begin with a scene-by-scene inventory of what carries the story. Note a whispered name, an alarm, a written message, a change of location, or a silent gesture. This is your own editorial review, not an automated accessibility score. For each item, ask which version of the film makes that information available and whether a viewer could miss it.
Review captions against the finished soundtrack
If you use automatic transcription, treat it as a draft. W3C WAI warns that automatic captions need accuracy checks and usually substantial correction. Review names, short words that reverse meaning, speaker changes, and sounds that matter to understanding the scene. Do not assume that a clean-looking subtitle file is accurate just because it imports successfully.
For an AI-assisted production, compare the text with what is actually audible in the final edit. A voice may have been regenerated after the script was approved; an editor may have moved or shortened a line. The script, the earlier transcript, and the finished soundtrack can therefore disagree. Keep the final audio as the reference for caption review.
Make one review pass for wording and a separate pass for timing. Watch a complete scene without stopping, then revisit cues that appeared too early, disappeared before you could read them, or competed with an important image. Keep a short list of corrections so a second reviewer can see which choices still need attention.
Plan visual description as its own creative task
W3C WAI describes several delivery approaches, including description integrated into a video’s audio and a separate described version or track. The appropriate choice depends on the content, available gaps in the soundtrack, and playback support. Captions alone do not explain a silent visual reveal.
Consider a fictional scene in which a character silently recognises a photograph. A description should make the relevant action understandable without inventing an emotion that the image does not establish. Decide what the viewer needs to know at that moment, rather than cataloguing every object in the frame. This example is illustrative and is not a report of a festival film.
Try the proposed description with the actual scene. If it covers dialogue or reveals an important detail before the intended moment, revise the wording or delivery. For a substantial project, seek a reviewer with relevant audio-description experience and involve people who use the described version. Their feedback can reveal gaps that a maker already familiar with the story may overlook.
Keep the version and delivery method explicit
Name the film version, caption language, and description version in your delivery notes. A folder containing several files called “final” invites mistakes. Use a consistent identifier shared by the video and its associated text files, and keep the earlier working exports separate from the handoff folder.
WebVTT is a W3C format for timed text tracks. That does not mean every screening system accepts it: ask the recipient which formats, language labels, and playback arrangements they support. If you provide captions embedded in the image as well as a separate track, label them clearly so the operator can avoid displaying two sets at once.
Do not change the edit after approving its caption and description files without rechecking them. Even a small cut near the beginning can shift later cues. Record the export you reviewed and the date of the check in the handoff notes.
Test the audience experience, then hand over the evidence
Open the exact delivery files in the intended player where possible. Check that the right language is selected, caption controls behave as expected, text remains readable, and the described version is easy to identify. A successful export is only one stage of delivery; the operator still needs to know which assets belong together.
Package the reviewed video, associated tracks or alternate versions, and a concise note explaining the available options. Include any unresolved playback limitation rather than silently assuming it will be fixed at the venue. This check is a practical production workflow, not a certification that the film or screening meets every accessibility requirement.
Milan AI Festival is a VIA Group event. Its programme and technical delivery requirements remain unannounced, so check the official festival information before preparing a submission. For a wider account of your production choices, pair these notes with our AI film project breakdown guide. Accessibility decisions deserve a place in that process explanation alongside generation, editing, and sound.