Reel builder

Reel Builder — Claude assembles the whole reel for you

The last step of the factory. Hand Claude a teardown (from reel-analyzer) or just an idea + the structure to follow, and it builds the finished reel — every beat scripted in your voice, with the on-screen text and the visual for each beat — so you can shoot/render it as-is.

When to use

You have the structure you want to model (or a hook + topic) and you want the entire reel assembled, not just notes. Run this after the analyzer + hook writer, or standalone with your own outline.

What you'll need

The teardown or structure to follow, your topic/offer, your voice, and the CTA (what you want them to do). Optional: your past winners (via ai-brain) so it sounds like you.

Instructions

You are my reel builder. Build a complete, ready-to-make reel.
STRUCTURE TO FOLLOW: [paste the reel-analyzer teardown, or describe the beat structure].
TOPIC: [what it's about] · VOICE: [yours] · GOAL/CTA: [save / follow / comment / DM].

Output a beat-by-beat build table — one row per beat:
| # | Spoken line (VO, my voice) | On-screen text | Visual / scene | ~secs |
Rules:
- Front-load the hook: beat 1's biggest element is moving on frame 0; the payoff lands in the first ~1.5s.
- The visual CHANGES every beat (show what's said — never plain text for a demo line).
- Original words on the modeled structure; total length matches the reference (±2s).
- End on ONE clear CTA that matches the deliverable.
Then give me: the final hook (3 options), the full VO script as one paragraph, and a caption + hashtags.

Pairs with: reel-analyzer (what to model) → viral-hook-writer (the opener) → this (the full build) → caption-and-hashtags (the post).

Rendering the reel — the video engine (make it, not just script it)

Scripting is half the job; this is how you turn the build table into a real, cinematic reel that actually gets watched. Use a hybrid engine, matched to the tier:

• Character-acting / cinematic beats → Seedance (image-to-video). Seedance runs on Runway via the RunwayML SDK with model: "seedance2" (portrait 720:1280 for a 9:16 reel, ~5s per clip). Generate a conditioning still of the character/scene, then i2v it so the subject genuinely MOVES and performs. Cost ≈ $6–12 per reel — reserve it for the premium ($599 done-for-you) tier, not the $20 self-serve one.
• Hook, chapter, text, payoff, CTA beats → Remotion code-render ($0). Fast, free, on-theme. This is the whole $20-tier reel and the text scaffolding of the premium one.

Non-negotiables (each of these was a real failure once — do not repeat them):

1. The hook needs a gripping VISUAL, not just text. Open on something that grabs (e.g. a creator hacking a glowing terminal, a lock cracking, sparks) — the payoff word lands in the first ~1.5s.
2. The character ACTS through a progressive action — never ping-pong/loop a few poses. Looping a small pose set back-and-forth (A→B→A→B) reads as "replaying" and plays actions in reverse. Each beat is ONE forward action, start to finish. Seedance i2v does this natively; posed-frame Remotion does NOT (it loops and reads choppy — avoid it for real acting).
3. Sync the visual to the VOICEOVER, not a wall-clock. Transcribe the VO (word timestamps) and cut/act ON the word — the action must match what's being said. Lead the cut ~10 frames.
4. Emphasis text must be high-contrast and readable on every background (e.g. bright amber #FFCC33 with a dark outline stroke reads on both dark and light beats).
5. Close on an on-screen CTA card, tied to the comment system. End with a themed "Comment [KEYWORD] and I'll DM you [the thing]" card (an IG-composer graphic that types the keyword) — use the SAME keyword the caption uses and that the auto-responder is wired to, so comment → auto-DM → lead actually fires. Wire the keyword in comment-responder BEFORE you post, so the moment a viewer comments the keyword they get the auto-DM. Honest only: the keyword must be really wired.
6. No fabricated numbers. If a metric can't be verified, don't put it on screen.

Verify before you ship — WATCH and MEASURE the playback, never trust a still. Extract frames at ~12fps and confirm: motion is progressive (no repeating/ping-pong autocorrelation period), no frozen beats while the VO talks, and each action's cut aligns to its narration cue. Then eyeball the whole thing end to end. A reel that was never watched ships broken.

Credentials: use an API key you were explicitly given, from your OWN project's env only. Never search the filesystem for a key or read another product's .env — if you don't have the key, stop and ask.

Honesty: only promise a CTA you can deliver (if "comment WORD → I'll DM it", actually send it).

---

Built by Ootto — the AI autopilot that connects your tools once and runs the busywork for you, automatically. Book a demo →