Skip to content
Back to search

lets-scroll

Agent skill that builds scroll-scrubbed "fly through the world" landing pages, AI-generated scenes + camera flights chained into one seamless continuous shot, driven by scroll. Claude Code / Codex / SKILL.md-compatible.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/aiwithhassan--lets-scroll/badge.svg)](/listings/aiwithhassan--lets-scroll)

HTML

<a href="/listings/aiwithhassan--lets-scroll"><img src="/api/listings/aiwithhassan--lets-scroll/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

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).

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 133 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # lets-scroll
  2. An agent skill — for Claude Code, Codex, and any `SKILL.md`-compatible agent — that
  3. turns any brand, product, or industry into a **scroll-driven cinematic landing page**.
  4. As the visitor scrolls, a pre-rendered camera travels through a series of AI-generated
  5. scenes as **one unbroken shot**: down flies it forward, up plays it back. The scroll
  6. position never triggers cuts or transitions — it simply moves time along a single
  7. continuous camera path, so the seams between scenes have to be perfect. Making them
  8. perfect (each clip is conditioned on its neighbour's actual rendered frames) is the
  9. core of what this skill does.
  10. You pick the flight's personality at the start — the skill implements it, it never
  11. re-decides it:
  12. - **Fly through the world** — the camera dives into each scene, pulls up and out, and
  13. hops across the miniature world to the next (the flagship diorama look).
  14. - **One continuous walkthrough** — a single forward flight that glides through each
  15. scene straight into the next, never pulling back; the seamless choice for grounded,
  16. photoreal art directions.
  17. - **Locked isometric glide** — one fixed camera angle for the whole film; the world
  18. slides past beneath the same view, no rotation, no reveals. The calmest and cheapest
  19. to re-roll.
  20. ## Install
  21. Copy the skill folder into your agent's skills directory:
  22. ```bash
  23. cp -R lets-scroll/skills/lets-scroll ~/.claude/skills/ # Claude Code
  24. cp -R lets-scroll/skills/lets-scroll ~/.codex/skills/ # Codex
  25. ```
  26. Then just ask for a scroll-through world landing page, or invoke `/lets-scroll`
  27. (`$lets-scroll` in Codex).
  28. ## Requirements
  29. - The [Monid CLI](https://monid.ai) with an API key and balance — the **default
  30. video-chain backend** (Seedance 2.0, billed per clip in USD; see below).
  31. - The [Higgsfield CLI](https://higgsfield.ai), authenticated (`higgsfield auth login`),
  32. with credits — renders the scene stills, the `kling3_0` fallback, and the whole
  33. chain when Monid is absent.
  34. - `ffmpeg` / `ffprobe` for frame extraction and encoding.
  35. - Python 3 with Pillow (for the mobile portrait canvases; also the optional
  36. transparent-scene knockout).
  37. - The [Codex CLI](https://github.com/openai/codex) (optional) — if present, the scene
  38. stills can be generated through Codex's built-in `image_gen` (the same GPT Image
  39. model), billed to a ChatGPT subscription instead of Higgsfield credits.
  40. - About the Monid default: verified 2026-07-25 — first/last-frame conditioning
  41. frame-locks, so it renders the full seamless chain; frames travel via Monid's
  42. free workspace file system. Pay-per-use with no subscription or monthly expiry
  43. (a 6-scene 1080p chain ≈ $27). The skill re-checks the endpoint schema each
  44. build and keeps qualification probes in the pipeline for when the catalog
  45. changes; Higgsfield credits remain the fallback biller.
  46. - **Manual asset path:** if you render the stills/clips yourself from the skill's
  47. prompt files, none of the AI CLIs above are needed — only `ffmpeg` (and Python
  48. for the optional knockout). The skill writes every prompt to a file plus a
  49. `HANDOFF.md` spec table per phase (prompt file, conditioning frame(s), the exact
  50. output filename, and a live status column), then **validates what comes back** —
  51. count, dimensions, aspect, duration, and that frame 0 of each clip matches the
  52. handed-over start frame — before anything is chained. Your video tool must accept
  53. a start frame (and, for fly-through connectors, an end frame) or the seams can't
  54. lock; the walkthrough architecture needs no end-frame support.
  55. ## What it does
  56. Two pipelines, one page. The art pipeline renders every scene still with GPT Image 2
  57. (via Higgsfield, or the Codex CLI on a ChatGPT subscription) under a shared style
  58. preamble, so the whole world reads as one place; the default look is a soft isometric
  59. clay diorama, with photoreal, papercraft, glossy-toy and neon-night as first-class
  60. alternates. The motion pipeline renders the camera chain itself with frame-locking
  61. video models — Seedance 2.0 via **Monid by default** (pay-per-clip USD), Seedance or
  62. Kling on Higgsfield credits as fallback; any model that can't hold a seam is
  63. disqualified. The page is a portable vanilla-JS scrub engine (blob-loaded, always
  64. seekable video) that drops into plain HTML, Next.js, Vue, or a server-rendered page —
  65. no stack assumptions.
  66. When invoked, the skill:
  67. 1. **Interviews you** — the subject/industry + pitch, a brand kit (import from a URL, hand
  68. it over, or have it proposed), art direction, the **camera style** (fly-through /
  69. continuous walkthrough / locked isometric glide — this decides the clip architecture),
  70. the **journey size** (2 scenes = a teaser, 4 = a short journey, 6 = the full film)
  71. and the ordered scenes the camera visits,
  72. the **asset source** (the skill renders everything via Monid/Higgsfield, or hands you
  73. every prompt + conditioning frame so you can render in your own tools and drop the
  74. files back in), whether you want the **mobile version** (a second chain rendered
  75. natively in 9:16 portrait — composed for phones, not a crop of the landscape film),
  76. and the **budget** —
  77. render tiers and stills source shown with estimated credit costs, approved before
  78. anything generates.
  79. 2. **Generates the assets** — one still per scene, then the camera chain. Fly-through:
  80. one "dive-in" clip per scene plus the **connector** clips that join consecutive
  81. scenes. Walkthrough / locked-iso: one forward **leg** per scene, no connectors at
  82. all — each leg starts on the previous leg's actual last frame. Either way every seam
  83. is frame-identical, because connectors/legs are conditioned on the **actual rendered
  84. frames** of their neighbours, never on a fresh render of the same scene.
  85. Mobile opt-in renders a parallel portrait chain the same way, frame-locked against its
  86. own 9:16 renders.
  87. 3. **Wires it up** — a config-driven scroll engine that plays the whole chain as one
  88. flight, serving the portrait clips and posters automatically on phones.
  89. ## What's in the skill
  90. ```
  91. skills/lets-scroll/
  92. ├── SKILL.md the procedure + the seam rule + gotchas
  93. └── references/
  94. ├── prompts.md intake checklist + every Higgsfield prompt template
  95. ├── pipeline.md copy-paste batch scripts (generate → frames → connectors → encode)
  96. ├── scrub-engine.js portable, config-driven scrub engine (blob-seek, lazy load, seam crossfade)
  97. ├── index-template.html a minimal standalone page that mounts the engine
  98. └── knockout.py background knockout for floating scenes
  99. ```
  100. ## Notes
  101. - Asset generation costs money (~N image gens on Higgsfield credits + N to 2N−1 video
  102. gens — walkthrough chains render N forward legs; fly-through chains render N dives +
  103. N−1 connectors — billed per clip on Monid by default; the mobile chain doubles the
  104. video gens)
  105. and takes a while — the skill runs generations in the background and polls. Monid
  106. pricing is per-token and printed per run; Higgsfield pricing isn't exposed by its
  107. CLI, so the skill calibrates against your live balance. Either way the estimated
  108. total is stated before spending.
  109. - Cheapest full pipeline test: a **2-scene teaser on the walkthrough architecture** —
  110. 2 stills + 2 sequential legs, no connectors, no end-frame support needed. The right
  111. first run, and a good live-demo format.
  112. - The generated `.mp4`/`.webp` assets are produced per project; they're not shipped here.
  113. ## License
  114. MIT — see [LICENSE](LICENSE).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.