leodavincs/kalem
leodavincs/kalem has 117 GitHub stars and 17 forks. GitHot measured +2 stars in 24 hours and +7 in 7 days. Last checked 2026-11-15 15:53 UTC. Last push 2026-11-06.
Measured star growth
- in 24h
- +2
- in 7 days
- +7
- in 15 days
- Update delayed
- in 1 month
- Update delayed
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
About
Claude Code skill: her sahnenin kendi kalemi var. Videoyu sahnelere böler, her sahneye uygun motoru seçer (Paper.js, PixiJS, HyperFrames, Three.js, Manim).
Creative & MediaREADME.en.md
Original
Kalem
Türkçe | English
Every scene has its own pen.
New: plate engine (levha)
Takes a machine designed in code (lever, pulley system, crank-piston, gear train), draws it like a Leonardo notebook plate, builds it stroke by stroke, runs its parts on their joints and turns it into a teaching video. The blue technical-drawing style needs no Codex and runs entirely in code.

Details and the one command: motorlar/levha
When you ask an AI to make a video with code, the results all look alike, because every scene is drawn with the same engine. Kalem ("pen" in Turkish) is a Claude Code skill that splits your video into scenes, looks at what happens inside each one and picks the engine that draws it best. It then renders each scene with that engine, joins them with transitions such as a page flip, and adds sound effects locked to the frame.
You describe what you want. The skill decides which pen draws which scene.

Which scene, which engine
| Inside the scene... | Engine |
|---|---|
| A line that draws itself, a sketch | Paper.js |
| A character, walking, one shape morphing into another | Plain canvas |
| Explosions, light, dust, thousands of particles | PixiJS |
| Text, numbers, counters, info cards | HyperFrames (HTML + GSAP) |
| Real 3D: orbiting camera, light, shadow | Three.js |
| Formulas, graphs, diagrams | Manim |
| A running machine as a notebook plate, teaching video | Levha (motorlar/levha) |
Helpers: Matter.js and Rapier (physics), gl-transitions (125 transitions incl. page flip), Tone.js (frame-locked sound effects), Anime.js, tsParticles, glTF-Transform.
The table reflects one person's taste: 20 tools raced on the same scene and were scored by eye. Run your own tests and change it. It lives in .claude/skills/kalem/SKILL.md. The skill and its prompts are written in Turkish; Claude follows them in any language.
Install
Needs macOS or Linux, Node 20+, Python 3, ffmpeg, Google Chrome and Claude Code. Manim is optional.
git clone https://github.com/leodavincs/kalem.git
cd kalem
./kurulum.sh
Use
Open Claude Code in the repo folder and describe your video:
claude
/kalem A 50-second landscape video about Leonardo da Vinci's flying machine...
Claude first shows the scene table (scene, engine, why) and waits for your approval, then renders every scene in its own folder, checks the frames and joins them into projeler/<name>/cikti.mp4.
Manual join: python3 birlestir.py projeler/<name> --gecis BookFlip --etiket (--etiket stamps a small card naming the engine on each scene).
How it works
- Time comes from the frame number; randomness is seeded, so the same command gives the same frames.
- Browser engines are captured frame by frame in Chrome (
ortak/yakala.py), not recorded in real time, so heavy scenes never stutter. - Each scene writes
olaylar.json(a line starts, something drops, a wing flaps). Sound is generated from that list, so it never drifts from the picture. - Scenes are independent and can be rendered in parallel.
License
MIT. Fonts in ortak/font/ are under the SIL Open Font License 1.1.
Leo (leodavincs)