--- name: "cinematic_trailer_production" description: "Cut a cinematic trailer for a film, series, game, or story: pick the placement and runtime first, log and score the moments that sell the story, draw a spoiler line and hold it, build the three-act trailer shape with a music-first method, place title cards and the button, then cut down to shorter variants by removing beats. Covers theatrical, online, teaser, and spot lengths, green and red band rules, sound design vocabulary, and delivery loudness. Triggers when someone asks to make, plan, restructure, or shorten a trailer or teaser, fix a trailer that spoils too much or feels flat, or decide trailer length for a placement." metadata: { "includeInPrompt": true } --- # Cinematic Trailer Production ## Purpose Cut a trailer that sells the story without spending it. A trailer is a promise: it tells the viewer what watching this will feel like, and it leaves the story itself unsold enough that the ticket still buys something. This skill is provider neutral. It works the same whether the footage is shot, generated, animated, or captured from a game, and whether the edit happens in Premiere, Resolve, CapCut, or ffmpeg. Neighbouring skills cover other ground: beat-synced-video-editing cuts picture to a music track, demo-video-production covers product demo trailers, and launch-tease-campaigns runs the social campaign around a launch. This skill is the trailer itself: selection, structure, sound, cards, and variants. Two failure modes account for most bad trailers. The first is the summary trailer, which retells the plot in order and gives away the ending. The second is the sizzle reel, all peak moments with no setup, so the viewer feels noise and remembers nothing. The method below avoids both: structure first, moments second, polish last. ## Runtime targets by placement Pick the placement before picking a single shot. Length is a contract with where the trailer will play. - Theatrical full trailer: 2:00 to 2:30. The Motion Picture Association caps theatrical trailers at 2:30, with one exception per studio per year. Aim at 2:20 so the rating card and billing fit inside the cap. - Online full trailer: 1:30 to 2:30. No formal cap online, but the same shape applies. Past 2:30, completion rates fall and the cut usually contains a second act 2. - Teaser: 0:30 to 1:00. Mood, one image, one line, a date or a season. Teasers sell a question, not a premise, and often ship before the work is finished, so they lean on tone and a single strong moment. - TV or streaming spot: 0:15, 0:30, or 0:60, cut to the slot exactly. A 15 second spot loses 3 to 5 of those seconds to its end card, so it carries one idea and nothing else. - Social cut: 0:15 to 0:45 vertical or square, built as its own cut (see variants), never a centre crop of the master with the title card sliced in half. Rating bands, for theatrical and public placement in the US: a green band trailer is cleared for the audience of the feature it accompanies and keeps language, violence, and sexual content inside advertising limits. A red band trailer may carry restricted material and may only run before R, NC-17, or unrated features, and behind age gates online. Decide the band before the selection pass, because it decides which shots are even eligible. ## Workflow 1. Write the one-line job. Before touching footage, write three things: the promise (what the viewer will feel), the audience (who this cut is for, since a horror cut and a family cut of the same film use different footage), and the one question the trailer leaves open. If the question can be answered from the trailer, the cut has already spoiled something. 2. Draw the spoiler line. Go through the source and mark the beats that stay protected: every reveal, reversal, and outcome from the final third, plus the climax wherever it sits. The working rule in film marketing is that material from the first two acts is fair game and the third act is not. Write the protected list down and pin it over the edit. Two hard rules: never show the climax, only the approach to it, and never use a shot whose meaning depends on a protected reveal, because viewers read more from a reaction face than editors expect. A trailer may reorder, compress, and put lines over images they were never spoken in. It may not invent a fake ending or promise a scene the work does not contain. 3. Log and score the selects. Watch the whole source and log every candidate moment with timecode, line or image, and a score from 1 to 5 on four axes: hook (works with no context), premise (teaches the story), stakes (shows what can be lost), and spectacle (pure size or beauty). Also flag eligibility: green band safe or not, dialogue clean or covered by music, works muted or not. A full trailer needs roughly 40 to 60 logged candidates to build 25 to 35 used shots. If logging feels slow, that is the job working: every trailer problem later is a selection problem now. 4. Pick the music before cutting picture. Music is the trailer's skeleton: its sections become the acts, its hits become the cuts, its stop-downs become the act turns. Choose one main cue with a quiet opening, a build, and a peak, or plan a two-cue build (setup cue, then a bigger second cue from the turn on). A licensed song can carry a trailer when its lyric or chorus states the theme, and trailer versions of songs often slow them down, strip them to piano, or add hits over the top. Whatever the source, map the cue first: mark every section change, hit, and silence on the timeline before a single shot is placed. 5. Build the radio cut. Assemble the dialogue and sound story alone, with no picture: the lines that set the premise, the turn line, the stakes lines, in order, over the music map. Play it for someone who has not seen the work. If they cannot say what the story is about and what question it leaves open, fix the radio cut, not the picture. This is the cheapest structural test in the whole process. 6. Lay the three-act build. Run `scripts/trailer_plan.py plan --seconds ` for the timing budget at your target length, then place shots to the music map inside it: - Cold open (about 5 percent): the single strongest self-contained moment, 3 to 8 seconds, no context needed. Logos and rating cards follow it, not precede it. - Act 1 (about 25 percent): the world, the lead, the tone, ending on the question the story answers. Shots run long here, 3 to 5 seconds. The audience must care before anything can be threatened. - The turn (one beat): a music drop, a shocking line, an image that reframes act 1. This is the hinge of the trailer. It gets silence or the biggest single hit so far, never a fade. - Act 2 (about 30 percent): the real premise, the conflict, the stakes rising. Cuts shorten to 1 to 2 seconds, the music builds, and each beat is worse or bigger than the last. - Act 3 (about 25 percent): a montage of the biggest images on the music peak, shots under a second at the top, ending mid-climax. Freeze or cut away on the approach, never on the landing. - Title card (3 to 7 seconds): the title, held long enough to read twice, on the cleanest sound in the trailer. - The button (3 to 6 seconds, full trailers only): one last beat after the title, a laugh, a scare, or a line that reopens the question. It is the exclamation point, and it is the first thing viewers quote. - Billing and date (2 to 3 seconds): billing block and release date. On spots, the date merges into the title card. 7. Run the sound design pass. Trailer sound has a small vocabulary, used with restraint: a hit accents a beat, a suckback pulls into one, a whoosh carries a transition, a riser builds across several seconds toward a reveal, and a drone beds a whole section in a mood. Working rules: save full-size hits for the turn, the act 3 peak, and the title. Give each act turn a stop-down, a drop to silence or near-silence, because silence is what makes the next hit land. Keep risers out from under dialogue lines that carry plot. After the first full pass, mute a third of the sound design and play it again; if the story reads clearer, keep the mutes. 8. Place the text cards. Cards carry what footage cannot: time jumps, stakes stated plainly, review quotes on later cuts. Rules: one idea per card, three to seven words, held long enough to read aloud at speaking pace (about 0.4 seconds per word, minimum 1.5 seconds). Never put a card over a face that is talking. Never let a card say what the next shot shows. Most strong trailers need three cards or fewer, and many need none. 9. Cut the variants by removing beats. A 60 or 30 second version is not the master played faster or trimmed everywhere. It is the same skeleton with whole beats removed: setup dies first, then secondary stakes beats, then half the act 3 montage. The cold open, the turn, the climax peak, the title, and the button survive longest. Run `scripts/trailer_plan.py variants --seconds ` for the cut-down budgets, and rebuild vertical social cuts from the selects log rather than cropping the master, reframing shot by shot so faces and the title stay whole. 10. Check the cut list. Export the timeline as a CSV with columns shot_id, act, duration_s, source_pos (the shot's position in the source as 0.0 to 1.0), and spoiler, then run `scripts/trailer_plan.py check --seconds `. It fails the cut on a missed runtime, an act outside its working share, a flagged spoiler, or a missing or unreadable title card, and it warns on every shot pulled from the final third of the source. Clear every fail and read every warning against the protected list from step 2 before picture lock. 11. Verify cold. Watch the locked cut once through, muted: the story and the escalation should still read. Watch once with eyes closed: the music shape and the lines should still build. Then show it to one person who knows nothing about the work and ask two questions: what is it about, and what happens at the end? The first answer tests the premise. The second tests the spoiler line: if they can answer it, the trailer gave away too much. ## Delivery specs - Online master: 1920x1080 or higher, 23.976 or 24 fps to match the source, loudness at -14 LUFS integrated with true peak at or below -1 dBTP, which is where YouTube normalises playback. - Broadcast: -23 LUFS under EBU R128 in Europe, -24 LKFS under ATSC A/85 in the US, true peak at or below -1 dBTP (EBU) or -2 dBTP (ATSC). Get the network's own sheet when one exists; it wins over these defaults. - Captions: supply a caption file for every public cut, including the teaser. Trailers get watched muted on social feeds, so burn nothing in, but ship the sidecar file. - Slate and naming: file names carry work, variant, band, and date, for example ferry_full_greenband_2026-10-18.mp4. Keep the selects log and the protected-beats list with the project; the next cut, often a review-quote trailer after release, starts from them. ## Output contract A finished trailer job delivers: - The master cut at the contracted placement length, plus every variant cut that was ordered, each named per the naming rule. - The selects log with scores, so any shot choice can be defended or swapped without rewatching the source. - The protected-beats list and the check script output showing the master passed with no unresolved spoiler warnings. - One line per variant stating what was removed, so a stakeholder asking for a missing scene gets an answer, not a re-edit. ## Operating rules - Structure before moments, and moments before polish. A trailer with the right shape and ordinary shots beats a trailer of great shots in no shape. - Sell the feeling, protect the ending. Any single shot can be cut if it crosses the spoiler line, including the best shot in the work. - Music first, picture second, sound design third, cards last. Work done out of that order gets redone. - Tone honesty is binding: a trailer cut to promise a comedy, a romance, or a horror film the work is not will sell one weekend and poison every review after it. - Cut variants by removing beats, never by speeding footage up or shaving every shot. A squeezed trailer feels panicked, not short. - No invented scenes, no fake endings, no dialogue edits that change what a character said. Reordering is craft; fabrication is lying to the buyer. - When a stakeholder note asks for a protected beat, show them the protected list and the cold-viewer test result, and offer the strongest unprotected alternative. Notes get answered, the line holds. Worked example: one invented thriller cut at 150, 60, and 30 seconds, with its spoiler audit and cut-down reasoning, is in `references/worked-cuts.md`.