Captions, Reframing & Deliverables
The film isn't done when the film is done. Festivals want subtitles, platforms want captions, social wants vertical — and AI turns that mountain of deliverables into a single afternoon pass. Here's how, and where you still have to look.
Nobody warns you about the second post-production. You lock picture, finish the mix, sit back — and then discover the film needs subtitles for the festival, captions for accessibility, a vertical cut for social, a square cut for the other social, three trailers' worth of clips, and an export in whatever codec this week's platform demands. Five years ago this was days of grind or a line item you couldn't afford. Today it's an afternoon — if you run it as one deliberate pass instead of a month of one-off panics.
The foundation is something you already built in Chapter 5: your transcript. Because AI transcription tied every spoken word to a timecode, captions and subtitles are now generated, not typed. DaVinci Resolve and Premiere Pro both create timed captions from your audio directly, and export them as sidecar files or burn them in. Festival screeners increasingly expect subtitles. Accessibility isn't optional — deaf and hard-of-hearing viewers are your audience too. And on social, most viewers watch with the sound off, which means on those platforms your captions are your dialogue track. One transcript feeds all three.
The one-pass deliverables workflow
Here's the pass I run once picture and mix are locked. Do it in this order and nothing gets forgotten:
- 1 · Generate captions from the locked cut. Auto-generate, then human-proof the entire file. AI reliably mangles names, places, and technical jargon — the exact words that make you look careless when they're wrong.
- 2 · Export caption formats per destination. Sidecar subtitle files for festivals and platforms that want them; styled burn-ins only where the destination demands it. Keep the clean master caption-free.
- 3 · Auto-reframe for vertical and square. Let the AI subject-tracker do the first pass on your 16:9 master, then review every single shot. More on this below.
- 4 · Run your export presets. Build a preset per destination once — festival master, streaming upload, social variants — and reuse them forever. Presets are how you stop re-solving solved problems at 2 a.m.
- 5 · Check everything off a written list. Every destination, every file, every spec, ticked on an actual checklist. Memory is not a deliverables strategy.
Step 3 deserves its warning label. Auto-reframing — the AI that pans a crop window around your 16:9 frame to track your subject for vertical — is genuinely good on singles and mediums, where there's one obvious subject to follow. On wides, the tracker loses the plot: two characters split the frame and it dithers between them, or it fixates on whoever's moving instead of whoever's talking. Composition was your decision at 16:9; it's still your decision at 9:16. Review every shot, and hand-keyframe the ones the machine got wrong.
The checklist mindset — and the marketing bridge
The deeper lesson of this chapter isn't any single tool; it's a mindset shift. Treat deliverables as a defined, finite pass with a written checklist, not a trickle of requests you handle as they arrive. The moment a festival emails asking for a different subtitle format, you want to be regenerating from your transcript in minutes, not rebuilding from scratch. This checklist thinking scales all the way up to professional delivery — DCPs, audio configurations, technical QC — and when your film is headed for real screens, the DCP & Delivery course in the Academy covers that end of the pipeline properly. Consider this chapter the on-ramp.
And notice what you're holding when the pass is done: a captioned master, verticals, squares, and a folder of clean clips. That's not just delivery — that's a marketing arsenal. The same transcript-driven tools that cut your film will pull quotable moments and cutdowns at a scale that used to require a social team. Promotion stops being a second production and becomes an extension of the finish. That bridge — from finished film to films people actually see — is where AI quietly does some of its best work for independent filmmakers.
A director I know auto-captioned her festival screener and shipped it without proofing. The film got in — and the programmer's congratulations email gently noted that her own lead character's name was spelled three different ways across ninety minutes. The AI had guessed phonetically, differently each time, and no human had read the file. Ten minutes of proofreading would have caught it. Auto-generate everything; proof-read everything. Both halves are the workflow.
Next chapter: the invisible fixes — AI object removal and shot cleanup, the tools that erase the boom pole, the logo you couldn't clear, and the stranger who wandered through your best take.
Recut is this course in tool form — an AI edit assistant that pulls selects from your footage, builds a first assembly, and preps cutdowns, ready to polish in your own editor.
