القائمة
Turn raw footage into a punchy vertical (9:16) short — hook, jump cuts, zoom punches, trending captions, beat-aware rhythm. Use for "make this a viral reel / short".
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Master motion-designer playbook for composing premium films with compose_motion — the decision process, optical composition, rhythm, restraint, transition craft, the physics of premium motion, worked examples with the reasoning exposed, and the failure modes to avoid. Read this BEFORE composing any motion piece (launch film, product reel, title sequence, brand sting).
Cut and animate to the music — get beats/onsets with analyze_audio, then land cuts and zoom/scale punches on the beat.
Plan and place B-roll cutaways over an A-roll talking track — from the media library or generated (BYOK).
The non-negotiable way to work — build every edit ON the Kaestral timeline via MCP tools so the user watches it happen, and render only with Kaestral's own Export. Read this before any editing task.
Apply modern short-form caption looks — word-pop, karaoke highlight, bold-boxed, typewriter — sized and placed for 9:16.
Give a clip or project a cohesive designed look — extract the palette, build a color grade, and choose on-brand type and captions.
| name | Viral-style reel |
| description | Turn raw footage into a punchy vertical (9:16) short — hook, jump cuts, zoom punches, trending captions, beat-aware rhythm. Use for "make this a viral reel / short". |
FIRST, follow the house rule in the build-in-kaestral skill: build everything ON the Kaestral timeline via tools (the user watches it happen), never render a standalone file, and finish with Kaestral's own Export. Every step below is a tool call — apply it, don't just describe it.
Your job: take what's on the timeline and turn it into a short-form vertical video that earns attention in the first 3 seconds and keeps it. Work in passes, visibly. Call get_timeline and get_media first so you know what you're working with.
Don't edit blind. Before cutting, actually look and listen:
see_video on the hero clip(s) (mode:"scene" to see distinct moments, or "interval" for an overview). You get real frames — read the subject, the framing, the action, and pick the strongest visual moment for the hook.get_transcript on the spoken clip → word-level timestamps of everything said. Now you know the actual hook line, the payoff, and where the filler is — so cuts and captions land on real content, not guesses.
Use this understanding throughout: hook = the best moment you SAW or the best line you READ; captions = the real words; b-roll = illustrate what's actually said.set_project_settings with aspectRatio: "9:16" and quality: "1080p". This is the single most important framing move for reels.apply_layout with full + fill so it covers the vertical frame (center-crop), or use set_clip_properties to scale/position the subject into frame. Keep faces/subjects in the upper-middle third.The opening 3s decides retention. Do at least one of:
move_clips (or split_clips + ripple_delete_ranges to cut the slow lead-in). Never open on dead air or "hey guys."add_texts a short, punchy line (3–6 words, e.g. "watch this till the end") at the top third, style it bold with a background box (see the caption-styles skill), animation popIn or wordPop. Keep it on screen ~1.5–2s.set_keyframes on scale) adds energy.Short-form lives on tight cuts — remove every pause, filler, and dead beat.
analyze_audio on the talking clip — it returns silenceRanges (dead-air, in project frames). Feed those ranges to ripple_delete_ranges to delete the pauses and close the gaps in sync. This is the highest-ROI edit for talking-head footage.remove_words (by transcript index) — it ripple-deletes the exact word ranges and closes the gaps, the word-accurate version of pause cutting.set_keyframes on scale from 1.0 to ~1.12 over 3–5 frames with interpolationOut: "smooth" (eases like a pro, not mechanical), then ease back. Punch in on reveals, punchlines, and the hook.set_clip_properties with speed — <1 for slow-mo on a peak moment (e.g. 0.5), >1 to blast through a slow stretch (e.g. 1.5–2). Both render in the export (video + pitch-corrected audio). Speed-ramp by splitting the clip and setting different speeds per segment.scale and position (smooth) for a slow drift so nothing sits static.beat-sync-cutting skill so cuts and punches land on the beat (uses analyze_audio).Most short-form is watched muted — captions are non-negotiable.
caption-styles skill. Use add_texts with a bold style + background box, animation wordReveal or highlightPop, placed in the lower-middle (safe above the UI). Keep 2–5 words per screen.generate_motion "Transition" stinger works. Auto-insert dissolves at cuts via the timeline's transitions action if the whole thing should feel smoother.apply_color a touch of contrast + saturation + a warm or cool push to unify the look (see creative-director). Extract the footage palette first with extract_palette so captions/graphics match.generate_title) with a CTA ("follow for part 2") works. Keep it under ~1.5s.export_project(mode:"video") — this renders the CURRENT TIMELINE you just built, through Kaestral. Do NOT render a separate file with ffmpeg. If the user would rather export themselves, tell them to press Export in the app.Tell the user which of these you applied and which need input from them (e.g. the spoken transcript for word-accurate caption timing).