lets-scroll

An agent skill — for Claude Code, Codex, and any SKILL.md-compatible agent — that turns any brand, product, or industry into a scroll-driven cinematic landing page. As the visitor scrolls, a pre-rendered camera travels through a series of AI-generated scenes as one unbroken shot: down flies it forward, up plays it back. The scroll position never triggers cuts or transitions — it simply moves time along a single continuous camera path, so the seams between scenes have to be perfect. Making them perfect (each clip is conditioned on its neighbour's actual rendered frames) is the core of what this skill does.

You pick the flight's personality at the start — the skill implements it, it never re-decides it:

• Fly through the world — the camera dives into each scene, pulls up and out, and hops across the miniature world to the next (the flagship diorama look).
• One continuous walkthrough — a single forward flight that glides through each scene straight into the next, never pulling back; the seamless choice for grounded, photoreal art directions.
• Locked isometric glide — one fixed camera angle for the whole film; the world slides past beneath the same view, no rotation, no reveals. The calmest and cheapest to re-roll.

Install

Copy the skill folder into your agent's skills directory:

cp -R lets-scroll/skills/lets-scroll ~/.claude/skills/   # Claude Code
cp -R lets-scroll/skills/lets-scroll ~/.codex/skills/    # Codex

Then just ask for a scroll-through world landing page, or invoke /lets-scroll ($lets-scroll in Codex).

Requirements

• The Monid CLI with an API key and balance — the default video-chain backend (Seedance 2.0, billed per clip in USD; see below).
• The Higgsfield CLI, authenticated (higgsfield auth login), with credits — renders the scene stills, the kling3_0 fallback, and the whole chain when Monid is absent.
• ffmpeg / ffprobe for frame extraction and encoding.
• Python 3 with Pillow (for the mobile portrait canvases; also the optional transparent-scene knockout).
• The Codex CLI (optional) — if present, the scene stills can be generated through Codex's built-in image_gen (the same GPT Image model), billed to a ChatGPT subscription instead of Higgsfield credits.
• About the Monid default: verified 2026-07-25 — first/last-frame conditioning frame-locks, so it renders the full seamless chain; frames travel via Monid's free workspace file system. Pay-per-use with no subscription or monthly expiry (a 6-scene 1080p chain ≈ $27). The skill re-checks the endpoint schema each build and keeps qualification probes in the pipeline for when the catalog changes; Higgsfield credits remain the fallback biller.
• Manual asset path: if you render the stills/clips yourself from the skill's prompt files, none of the AI CLIs above are needed — only ffmpeg (and Python for the optional knockout). The skill writes every prompt to a file plus a HANDOFF.md spec table per phase (prompt file, conditioning frame(s), the exact output filename, and a live status column), then validates what comes back — count, dimensions, aspect, duration, and that frame 0 of each clip matches the handed-over start frame — before anything is chained. Your video tool must accept a start frame (and, for fly-through connectors, an end frame) or the seams can't lock; the walkthrough architecture needs no end-frame support.

What it does

Two pipelines, one page. The art pipeline renders every scene still with GPT Image 2 (via Higgsfield, or the Codex CLI on a ChatGPT subscription) under a shared style preamble, so the whole world reads as one place; the default look is a soft isometric clay diorama, with photoreal, papercraft, glossy-toy and neon-night as first-class alternates. The motion pipeline renders the camera chain itself with frame-locking video models — Seedance 2.0 via Monid by default (pay-per-clip USD), Seedance or Kling on Higgsfield credits as fallback; any model that can't hold a seam is disqualified. The page is a portable vanilla-JS scrub engine (blob-loaded, always seekable video) that drops into plain HTML, Next.js, Vue, or a server-rendered page — no stack assumptions.

When invoked, the skill:

1. Interviews you — the subject/industry + pitch, a brand kit (import from a URL, hand it over, or have it proposed), art direction, the camera style (fly-through / continuous walkthrough / locked isometric glide — this decides the clip architecture), the journey size (2 scenes = a teaser, 4 = a short journey, 6 = the full film) and the ordered scenes the camera visits, the asset source (the skill renders everything via Monid/Higgsfield, or hands you every prompt + conditioning frame so you can render in your own tools and drop the files back in), whether you want the mobile version (a second chain rendered natively in 9:16 portrait — composed for phones, not a crop of the landscape film), and the budget — render tiers and stills source shown with estimated credit costs, approved before anything generates.
2. Generates the assets — one still per scene, then the camera chain. Fly-through: one "dive-in" clip per scene plus the connector clips that join consecutive scenes. Walkthrough / locked-iso: one forward leg per scene, no connectors at all — each leg starts on the previous leg's actual last frame. Either way every seam is frame-identical, because connectors/legs are conditioned on the actual rendered frames of their neighbours, never on a fresh render of the same scene. Mobile opt-in renders a parallel portrait chain the same way, frame-locked against its own 9:16 renders.
3. Wires it up — a config-driven scroll engine that plays the whole chain as one flight, serving the portrait clips and posters automatically on phones.

What's in the skill

skills/lets-scroll/
├── SKILL.md                    the procedure + the seam rule + gotchas
└── references/
    ├── prompts.md              intake checklist + every Higgsfield prompt template
    ├── pipeline.md             copy-paste batch scripts (generate → frames → connectors → encode)
    ├── scrub-engine.js         portable, config-driven scrub engine (blob-seek, lazy load, seam crossfade)
    ├── index-template.html     a minimal standalone page that mounts the engine
    └── knockout.py             background knockout for floating scenes

Notes

• Asset generation costs money (~N image gens on Higgsfield credits + N to 2N−1 video gens — walkthrough chains render N forward legs; fly-through chains render N dives + N−1 connectors — billed per clip on Monid by default; the mobile chain doubles the video gens) and takes a while — the skill runs generations in the background and polls. Monid pricing is per-token and printed per run; Higgsfield pricing isn't exposed by its CLI, so the skill calibrates against your live balance. Either way the estimated total is stated before spending.
• Cheapest full pipeline test: a 2-scene teaser on the walkthrough architecture — 2 stills + 2 sequential legs, no connectors, no end-frame support needed. The right first run, and a good live-demo format.
• The generated .mp4/.webp assets are produced per project; they're not shipped here.

License

MIT — see LICENSE.