paper-collage-video
Make paper-collage ("Vox-style") explainer videos and kinetic lyric videos with no API keys and no paid services: torn colour paper, scissor-cut photo stickers, tape, halftone dots and words slapped on in the beat, while a camera travels across one big scrapbook wall. Uses Remotion plus free local Python tools (rembg cut-outs, librosa beat detection, optional faster-whisper timing). Use this skill whenever the user wants a collage, cut-paper, scrapbook, zine, editorial or Vox-like explainer, a lyric or lyrical video, a kinetic-typography music video, or a "video from my song / my script / my photos" in that look, in any language (Hindi, English, Hinglish, Urdu, ...), even if they never say "collage" or "Remotion".
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/satyasanatan989--paper-collage-video)HTML
<a href="/listings/satyasanatan989--paper-collage-video"><img src="/api/listings/satyasanatan989--paper-collage-video/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
You need **Node.js 18+**, **ffmpeg** and **Python 3.9+**.
git clone https://github.com/satyasanatan989/paper-collage-video.git pip install numpy pillow scipy librosa soundfile # required pip install rembg onnxruntime faster-whisper # optional: photo cut-outs, voice timing python paper-collage-video/scripts/check_setup.py
Then give it to your agent: - **Claude Code**: copy the folder to `~/.claude/skills/paper-collage-video/` (all projects) or `.claude/skills/` in one project. - **Google Antigravity**: copy the folder to `~/.gemini/antigravity/skills/paper-collage-video/` (all workspaces) or `<your workspace>/.agents/skills/paper-collage-video/` (one workspace), then restart Antigravity. - **Codex / other agents**: keep the folder in your project; the agent reads `AGENTS.md`.
**Try it without an agent:**
cd paper-collage-video/template npm install npx remotion studio # preview and scrub npx remotion render Collage renders/demo.mp4 # render the leap-year demo
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.
SKILL.md · Manifest · 95 lines
- ---
- name: paper-collage-video
- description: >
- Make paper-collage ("Vox-style") explainer videos and kinetic lyric videos with no API keys and no paid
- services: torn colour paper, scissor-cut photo stickers, tape, halftone dots and words slapped on in the
- beat, while a camera travels across one big scrapbook wall. Uses Remotion plus free local Python tools
- (rembg cut-outs, librosa beat detection, optional faster-whisper timing). Use this skill whenever the user
- wants a collage, cut-paper, scrapbook, zine, editorial or Vox-like explainer, a lyric or lyrical video, a
- kinetic-typography music video, or a "video from my song / my script / my photos" in that look, in any
- language (Hindi, English, Hinglish, Urdu, ...), even if they never say "collage" or "Remotion".
- ---
- # Paper-collage video (explainer + lyric), no API keys
- You build the whole video locally. Nothing in this skill needs an API key, a login or credits: images come
- from the user's own photos, frames of their own videos, public-domain scans, or are drawn in code; the
- audio is the user's own song or voice-over. Say this up front, because it is the reason people pick this skill.
- The look: torn colour sheets, cut-out photo stickers with a cream scissor margin, words on torn paper strips,
- tape, halftone dots. Everything lands with a "slap" on the beat and stays stuck to the wall. The camera
- whips from one patch of the wall to the next, one patch per line or idea, and the ending pulls back to
- reveal the whole scrapbook. This continuity is what keeps it from feeling like a slideshow.
- ## What you need on the machine
- Run `python scripts/check_setup.py`. It checks Node 18+, ffmpeg, and the Python packages, and prints the
- exact install command for anything missing. Install only what is missing, and ask before installing.
- `rembg` downloads its cut-out model once (about 170 MB, from GitHub, no account).
- ## Workflow
- Copy `template/` to a new project folder next to the user's files (for example `my-video/`), run
- `npm install` there, and work in that copy. Never edit the skill's own template.
- 1. **Collect inputs, in one message, only what is missing.**
- - Kind: explainer (voice-over or text) or lyric video (song).
- - Audio: the song section or voice-over file. Songs: only ones the user owns or has the right to use.
- - Words: the lyrics or the script, from the user. Do not write song lyrics from memory.
- - Format: 9:16 (Reels/Shorts, default) or 16:9. Language/script of the on-screen words.
- - Pictures (optional): their photos, a video to take frames from, or "draw it in code".
- 2. **Map the audio.** `python scripts/beats.py song.mp3 --start 37.2 --end 82.2 --out <project>`
- cuts the section to `public/audio.mp3` and writes `src/beats.json` (beat times from 0).
- Explainer with voice-over: `python scripts/transcribe.py vo.mp3 --lang en --out <project>` writes
- word times to `words.json` (runs locally, no key). For Hindi/Urdu songs whisper is unreliable: ask for
- an `.lrc` file or line start times, or estimate from beats and say it is an estimate.
- 3. **Write the line map**: each lyric line or script sentence gets a start time, snapped to the nearest
- beat when it is within about 0.12 s. Show it to the user as a short table before building.
- 4. **Pick a palette and plan the wall.** Read `references/style-guide.md`. One station (one patch of wall)
- per line/idea, 5-12 stations for 30-60 s. Show a rundown table (time, words, what we see) and wait for a yes.
- Changing a table is cheap; changing code is not.
- 5. **Make the pictures.** List frames and photos in a JSON for `scripts/cutout.py` (stickers and torn
- scraps, see `references/cutouts.md`). Run it and look at its contact sheet before using anything:
- remove-background mistakes are common and easy to spot there. No pictures? Use the code primitives
- (disc, heart, clock, calendar, icon, burst, orbit, sprinkle) instead.
- 6. **Textures**: `python scripts/textures.py --format 9:16 --out <project>` (palette from the storyboard).
- 7. **Write `src/storyboard.json`.** The schema and every element type are in
- `references/storyboard-schema.md`; working examples are in `examples/`. Words, pictures and props are
- timed by `beat` (n-th beat from the station start) or `at` (seconds), so they hit the music.
- 8. **Review stills, not the whole video.** `python scripts/review.py <project> 2 6.5 11 ...` renders those
- seconds and tiles them into `review/sheet.jpg`. Look at it like a director: words cut off or
- overlapping, faces covered, a station that is empty for more than a beat, text over a busy area. Fix, re-shoot.
- 9. **Render**: `npx remotion render Collage renders/video.mp4 --crf=19` inside the project.
- Check the duration and that audio is present (`ffprobe`). Tell the user what you checked and what you
- did not (you cannot listen to the sync yourself; ask them to watch with sound).
- ## Rules that make it look right (and why)
- - **One wall, never slides.** Things stay stuck after they land and the camera moves to them. Fading one
- card out and another in is the slideshow look this style exists to avoid.
- - **Literal pictures.** Show what the words say (a calendar tearing for "every moment", a heart for "heart").
- Viewers connect with literal images; clever metaphors make them scroll.
- - **Words are paper.** One word or short phrase per strip, 110-220 px tall on a 1080-wide frame, rotated
- a few degrees, never smaller than 70 px (people watch on phones). At most two levels of text per station.
- - **Hit the beat, not every beat.** Words land on the beats where they are sung or spoken; props land on
- the beats in between. Something new every 1-3 s, faster in a chorus.
- - **Contrast.** Strips are solid paper, so text is always readable. Keep the bottom 15% of a 9:16 frame
- free of words (platform buttons sit there).
- - **Deterministic.** No `Math.random()` or clock reads in the composition; the engine uses seeded noise, so
- every render matches the preview.
- - **Credit and disclose.** Remind the user to credit the song and artists. If any picture came from an AI
- generator, they should say so when posting. Do not put another brand's logo or name on screen; "Vox-style"
- describes a look, it is not a brand to imitate.
- ## Files
- | File | Read when |
- |---|---|
- | `references/storyboard-schema.md` | writing or editing `storyboard.json` (all element types and options) |
- | `references/style-guide.md` | choosing palette, fonts, layouts, transitions for a new video |
- | `references/lyric-video.md` | the user brings a song |
- | `references/explainer.md` | the user brings a topic, script or voice-over |
- | `references/cutouts.md` | turning photos or video frames into stickers and torn scraps |
- | `references/troubleshooting.md` | something fails (Windows encoding, whisper/av errors, fonts, slow renders) |
- | `examples/leap-year-explainer.json` | a complete code-only explainer (it is the template's demo) |
- | `examples/lyric-video.json` | a lyric video built from cut-outs |
README.md · README · 123 lines
- <div align="center">
- # 📰✂️ Paper Collage Video
- ### Vox-style explainers & lyric videos with your AI agent. **No API key. No credits. 100% local.**
- <p>
- <img src="https://img.shields.io/badge/API_keys-none-2ea44f?style=for-the-badge" alt="No API keys">
- <img src="https://img.shields.io/badge/Agent_Skill-Claude_Code_·_Antigravity_·_Codex-7c3aed?style=for-the-badge" alt="Agent skill">
- <img src="https://img.shields.io/badge/Rendered_with-Remotion-0B84F3?style=for-the-badge" alt="Remotion">
- <img src="https://img.shields.io/badge/License-MIT-yellow?style=for-the-badge" alt="MIT">
- </p>
- <table>
- <tr>
- <td align="center"><img src="assets/lyric-showcase.gif" width="300" alt="Hindi lyric video made with this skill"><br><b>Lyric video</b> (Hindi)</td>
- <td align="center"><img src="assets/demo-leap-year.gif" width="300" alt="Code-only explainer demo"><br><b>Explainer</b> (code only, zero pictures)</td>
- </tr>
- </table>
- **Type one sentence → get a finished, beat-synced paper-collage video.**
- [▶ Full demo with sound](assets/demo-leap-year.mp4) · [Install](#-install-2-minutes) · [How it works](#-how-it-works) · [Examples](examples/)
- </div>
- ---
- ## 😩 The problem
- Collage explainers and kinetic lyric videos are everywhere on Reels, Shorts and TikTok. Making one means
- After Effects skills, or chaining paid AI tools that **eat credits and need API keys**.
- ## ✨ The fix
- An open **agent skill**: you ask Claude Code (or Antigravity, Codex, or any agent that reads `SKILL.md`) in plain words,
- and it builds the video **on your own computer**:
- - ✂️ **Torn paper, scissor-cut stickers, tape, halftone dots**: the editorial collage look
- - 🥁 **On the beat**: every word and picture slaps down on a detected beat of *your* song or voice-over
- - 🎥 **Not a slideshow**: one scrapbook wall, the camera whips between patches, then pulls back to reveal it all
- - 🌏 **Any language**: Hindi / Devanagari and English fonts bundled, any Fontsource font for other scripts
- - 🖼️ **Your photos, your video frames, or none at all**: built-in props (sun, moon, heart, clock, calendar,
- lamps, marigolds, stars, confetti) make a full video from pure code
- - 💸 **$0**: Remotion + free Python tools (librosa, rembg, faster-whisper). Nothing calls a paid API.
- ## 💬 Just ask
- > "Make a 45-second paper-collage lyric video from `song.mp3`, 0:37 to 1:22. Here are the lyrics: …"
- > "Make a Vox-style explainer on why we have leap years, 9:16, with my voice-over `vo.m4a`."
- > "Turn these 5 photos into a scrapbook Reel synced to `track.mp3`."
- > "मेरे गाने का कोलाज लिरिकल वीडियो बनाओ" (works in Hinglish / Hindi too)
- The agent asks for anything missing, shows a short rundown for you to approve, builds it, checks frames, and
- renders `renders/video.mp4`.
- ## 🚀 Install (2 minutes)
- You need **Node.js 18+**, **ffmpeg** and **Python 3.9+**.
- ```bash
- git clone https://github.com/satyasanatan989/paper-collage-video.git
- pip install numpy pillow scipy librosa soundfile # required
- pip install rembg onnxruntime faster-whisper # optional: photo cut-outs, voice timing
- python paper-collage-video/scripts/check_setup.py
- ```
- Then give it to your agent:
- - **Claude Code**: copy the folder to `~/.claude/skills/paper-collage-video/` (all projects) or `.claude/skills/` in one project.
- - **Google Antigravity**: copy the folder to `~/.gemini/antigravity/skills/paper-collage-video/` (all workspaces)
- or `<your workspace>/.agents/skills/paper-collage-video/` (one workspace), then restart Antigravity.
- - **Codex / other agents**: keep the folder in your project; the agent reads `AGENTS.md`.
- **Try it without an agent:**
- ```bash
- cd paper-collage-video/template
- npm install
- npx remotion studio # preview and scrub
- npx remotion render Collage renders/demo.mp4 # render the leap-year demo
- ```
- ## 🧠 How it works
- ```
- your song / voice-over ─▶ beats.py (beat grid) ─▶ transcribe.py (word times, optional, offline)
- your photos / frames ─▶ cutout.py (stickers + torn scraps, local rembg)
- your palette ─▶ textures.py (torn sheets, word strips, newsprint wall)
- │
- ▼
- src/storyboard.json (stations · words · pictures · props, each on a beat)
- │
- ▼
- Remotion engine (Collage.tsx) ─▶ renders/video.mp4
- ```
- | Folder | What's inside |
- |---|---|
- | [`SKILL.md`](SKILL.md) | the workflow your agent follows |
- | [`references/`](references/) | storyboard schema, style guide, lyric & explainer guides, cut-outs, troubleshooting |
- | [`scripts/`](scripts/) | setup check, beats, transcription, cut-outs, textures, review contact sheet, demo audio |
- | [`template/`](template/) | the Remotion project copied for every new video |
- | [`examples/`](examples/) | a code-only explainer and a full lyric-video storyboard |
- ## 🆚 Why not the API-based tools?
- | | This skill | Typical AI video pipelines |
- |---|---|---|
- | API keys | **none** | 1-3 paid keys |
- | Cost per video | **$0** | credits per image, clip, voice, song |
- | Sync to music | **beat-detected, frame-exact** | approximate |
- | Same result every render | **yes (deterministic)** | no |
- | Edit afterwards | **change a JSON line, re-render** | regenerate and pay again |
- ## 🙏 Good practice
- - Use songs you own or have the rights to, and credit the artists when you post.
- - Use your own photos, or public-domain images. The lyric showcase art above was made from the author's own
- AI-generated clips (Google Flow); if you use AI images, say so when posting.
- - "Vox-style" describes a look. Don't put another brand's logo or name in your videos.
- - Remotion is free for individuals and small teams; bigger companies need a [Remotion licence](https://www.remotion.dev/license).
- ## ⭐ Like it?
- If this saved you credits, **star the repo** so other creators can find it, and share what you make.
- Issues and pull requests (new props, fonts, palettes, languages) are very welcome.
- ## 📄 Licence
- MIT, see [`LICENSE`](LICENSE). Fonts are under the SIL Open Font License via Fontsource.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.