To caption an AI short film, work from the finished soundtrack, correct every spoken line, add the sounds needed to follow the story, and test the captions against the exported film. Treat this as an editing pass with its own review time. A transcript generated in seconds is a starting point, not a finished delivery.

The workflow below is our editorial recommendation for a small film team. It includes an invented scene to make the decisions concrete; it is not an official Prompt2Screen submission specification.

Decide what the audience needs to read

Captions communicate speech and meaningful audio information to viewers who cannot hear it. W3C distinguishes these from translated subtitles, while noting that terminology varies between regions. Automatic captions need accuracy checks and correction before they are ready for viewers. See W3C’s captions guidance.

For your project, write a short delivery note before opening a caption tool: the spoken language, the intended audience, the language of each text track, and where the film will play. A teammate should be able to tell which version they are reviewing without asking you.

If you want Italian and English versions, assign a separate review to each. Keep character names and invented place names in a shared spelling list. Ask a fluent reviewer to check translated dialogue against the scene, especially when a joke, hesitation or double meaning matters.

Create one reference cut for the caption pass

Export the version you intend to caption and give it a clear revision number. Our suggested naming pattern is film-title_cut-04, with the same revision on the caption file. Keep a note of who owns the next correction.

Imagine an AI short about a night attendant in an empty museum. The attendant says, “Leave the east door open.” A lock clicks somewhere off screen. A second voice answers, “It already is.” Those three moments form the reference sequence for this example.

Listen to the exported audio rather than relying on the script you prompted. If the selected voice take says “side door” instead of “east door,” decide whether to fix the soundtrack or preserve the spoken wording. A caption should not quietly repair a production mistake that remains audible.

Once caption review begins, record any edit that changes timing. Replacing a shot with a longer generation or adding an opening title means the old text track needs another synchronization check.

Caption the clue without explaining the plot

W3C’s transcription guidance recommends accurate speech, relevant speaker identification and meaningful non-speech sound. It also recommends splitting text at logical phrases.

In our museum scene, an example cue is [lock clicks]. If the source of the sound is deliberately hidden, [intruder unlocks the east door] would reveal information the audience has not received. Describe what is audible at that moment.

Likewise, a label such as UNSEEN VOICE: may be more appropriate than naming the speaker before the film reveals them. The choice depends on what the scene has already established. Keep a separate list of these creative decisions so a later translator does not accidentally undo them.

Review the film’s sound cues with the editor. Our sound-design workflow can help identify which effects carry an action or change the audience’s understanding. This is particularly useful for a film with little dialogue: an important sound can still tell part of the story.

Test the timing through the dramatic beat

Run the museum exchange without stopping. Does the second voice’s reply appear before the lock clicks? Does the text reveal the answer while the attendant is still waiting? Adjust the cue so the reader encounters the reply with the performance.

Then review the same passage for reading comfort. If you spend the entire shot reading, try a clearer phrase division and watch again. Preserve the meaning and the sound’s timing. Avoid compressing a character’s line into a different statement just to fit a crowded frame.

Use the most demanding moment as your layout test: perhaps a bright gallery wall, a face near the lower edge, or a clue written on a label. Check whether the caption obscures something the story needs. Repeat the check on the intended viewing screen, since a comfortable editor preview may hide a problem at smaller sizes.

Choose a delivery method deliberately

Adobe’s current Premiere export documentation describes burned-in, sidecar and embedded caption exports. A sidecar is a separate caption file; burned-in text becomes part of the picture. Embedded options depend on supported formats.

Ask the recipient which method and file format their playback system accepts. Keep an editable caption source alongside the picture master so corrections do not require rebuilding the text from scratch.

For the museum film, your handoff note could read: “Cut 04; Italian dialogue; Italian captions reviewed; English translation awaiting review.” Only mark a track reviewed after someone has actually checked it. Put unfinished versions in a separate working folder so they cannot be mistaken for the screening copy.

Watch the delivered file from beginning to end

Open the actual export in the intended player. For a separate track, load it explicitly. Check the first line, a middle exchange and the final cue, then watch the whole short without interruption. If all cues are offset, investigate the shared start point; if the mismatch grows or changes suddenly, revisit the timing and edit history.

Ask a fresh viewer to note any moment they could not follow while using the captions. Our guide to useful film feedback offers a simple way to collect specific observations.

Captions address the soundtrack. A film may also need audio description or a descriptive transcript to communicate essential visuals; W3C’s media planning guide explains those separate needs.

Rehearse before the next film challenge

Try this process on a short practice scene before your next event. Time the correction, export and playback checks, then reserve that time in your production plan.

VIA’s Prompt2Screen takes place on 10 October 2026 at Blox Space in Torino, according to the official event and registration page. Check that page and the organizer’s brief for current delivery instructions. Bring a caption workflow you have rehearsed, and adapt the final files to the requirements provided on the day.

Sources & further reading

Updated 23 September 2026. Event availability and final details are maintained on the linked official event pages.

Your next scene starts here.

Join the creators and the audience at Prompt2Screen Torino.

Find your pass ↗