manim
/manim
Tworzy film wyjaśniający w stylu Manim z narracją: skrypt, animację, lektor ElevenLabs, napisy i render 1080p.
- Autor
- Alexzcoder ↗
- Repozytorium
- manim-skill ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 11.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 6
- Skrypty
- 5
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 11.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- 5 skryptów w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: fetch_clips.sh, narrated.py, review.sh, speed.sh, watch_render.sh
Co robi i kiedy
Tworzy film wyjaśniający w stylu Manim z narracją: skrypt, animację, lektor ElevenLabs, napisy i render 1080p.
Wyzwalacze z opisu
Oryginalny opis (EN)
Make a narrated explainer video with Manim (3Blue1Brown style): script it, animate it, add an ElevenLabs voice and phrase-by-phrase captions, check the layout, render 1080p, and deliver the MP4. Use when the user asks for an explainer/animated video, says "make a video about X", "manim video", "animate this", "turn this doc into a video", or types /manim <topic or source files>.
Jak działa
- Rules that hold for every video
- Steps
- Gotchas learned the hard way
Podgląd SKILL.md
---
name: manim
description: |
Make a narrated explainer video with Manim (3Blue1Brown style): script it, animate it,
add an ElevenLabs voice and phrase-by-phrase captions, check the layout, render 1080p,
and deliver the MP4. Use when the user asks for an explainer/animated video, says
"make a video about X", "manim video", "animate this", "turn this doc into a video",
or types /manim <topic or source files>.
---
# manim: narrated explainer videos
Input: a topic, and usually source material (PDF, doc, repo, data). Output: a 1080p MP4 with
voice and captions in the user's Downloads folder, plus the scene file so it can be re-rendered.
Files next to this SKILL.md (below, `$SKILL` means this skill's folder: `~/.claude/skills/manim` for a
manual install, or the plugin's `skills/manim` folder):
- `narrated.py`: `Narrated` scene base class. `with self.say("line"):` syncs a beat to its
narration clip and draws the captions. Copy it next to the scene file.
- `fetch_clips.sh`: downloads narration clips from signed URLs.
- `review.sh`: contact sheets + duration + audio check for a rendered video.
- `speed.sh`: speed-up (default 1.75x) with pitch preserved.
- `watch_render.sh`: live progress bar for a running render (for the user to run).
## Rules that hold for every video
- **No invented data.** Every number on screen must trace to the user's sources. Verify them
(read the PDFs' text *and* their charts). Anything illustrative, schematic or stylised is
labelled so on screen.
- **Show the subject, not only charts.** If the topic has real geometry or objects (routes,
maps, networks, machines), look for the real data (GTFS shapes, coordinates, the user's
repo) and animate that. A video of bar charts alone is the failure mode.
- **No LaTeX.** Use `label()` / `Text` only. Never `MathTex`, `Tex`, `DecimalNumber`, or
axis `include_numbers` (draw tick labels by hand with `label`).
- One idea per beat; one `say()` per beat; `self.wipe()` between sections.
- Keep scene content above y = -3.05; the caption line lives below that.
- Narration text spells numbers as words ("seventy seven"); on-screen text uses digits.
## Steps
1. **Understand the material.** Read the sources. Write the outline: 8-14 sections, each with
its message and the picture that carries it. Target ~10 chars of narration per second;
7-11 minutes is ~4,500-6,500 characters. Proceed without asking unless the topic is unclear.
2. **Set up** (project folder, once):
```bash
uv venv -q --python 3.12 .venv && uv pip install -q --python .venv/bin/python manim
cp $SKILL/narrated.py .
```
Needs `ffmpeg` on PATH.
3. **Write the scene file**: one class deriving from `Narrated`, one `part_*` method per
section, `construct` calls them with `self.wipe()` between. Put the hard-coded figures in
one block at the top.
4. **Preview without voice and fix the layout.** Captions are timed by word count here.
```bash
.venv/bin/manim -ql --disable_caching scene.py MyVideo
$SKILL/review.sh media/videos/scene/480p15/MyVideo.mp4 <scratch>/sheets
```
Read every contact sheet. Fix text off-frame, overlaps, labels under lines, anything in
the caption band. Re-render and re-check the frames you fixed. To check one section fast,
render a subclass whose `construct` calls only that `part_*`, with `-s` for the last frame.
5. **Narration** (spends the user's ElevenLabs credits, ~1 credit per character; say the
total before generating if it is large, and never regenerate a clip that succeeded):
- Dump the lines: `NARRATION_DUMP=narration.json .venv/bin/manim -ql --dry_run scene.py MyVideo`
- ElevenLabs MCP: `creative_list_voices` to pick a clear narrator voice (don't ask),
`creative_create_flow` once, then one `creative_generate_speech` per line with
`generations_count: 1`, the same `flow_id` and `voice_id`, model `eleven_v4`.
- `creative_get_flow_run_status` with up to 20 `session_ids` per call until
`all_completed`. Match results to lines by their `prompt` text, not by order.
- Download with the signed `content_url` (the signature is required):
`$SKILL/fetch_clips.sh audio <<< "01 <url>"` (one `NN URL` per line).
Every row must print `200` and a duration; the clip count must equal the line count.
- No ElevenLabs connector: deliver the captions-only version and say so.
- If narration lines change later, regenerate only the changed clips, keeping numbering.
6. **Final render and checks.**
```bash
NARRATION_AUDIO_DIR=audio .venv/bin/manim -ql --disable_caching scene.py MyVideo # preview
NARRATION_AUDIO_DIR=audio .venv/bin/manim -qh --disable_caching scene.py MyVideo # 1080p60, run in background
$SKILL/review.sh media/videos/scene/1080p60/MyVideo.mp4 <scratch>/final
```
Send the `-qh` render's output to a log file. Get the animation count from a `--dry_run`
with the same audio dir ("Played N animations"), then give the user this to run in a
second terminal for a live bar with time left: `$SKILL/watch_render.sh render.log N`.
When they ask how far along it is, report `grep -c "movie file written" render.log` of N.
Confirm: no errors in the log, duration in range, audio present (mean volume around
-28 dB), no pause longer than ~2 s, spot-check frames. A 10-minute video takes roughly
10-15 minutes to render at `-qh`.
7. **Deliver.** Copy the MP4 to the user's Downloads folder (on WSL:
`/mnt/c/Users/<user>/Downloads/`) and verify the copy with `cmp`. If asked for a faster
cut: `$SKILL/speed.sh in.mp4 out.mp4 1.75` and deliver both.
Report: paths, duration, credits spent, what was checked, and what was not (you cannot
listen to the audio, so pronunciation is unverified).
## Gotchas learned the hard way
- Thousands of small mobjects (e.g. a dot per bus) make Cairo crawl. Merge them into a few
multi-subpath `VMobject`s with `append_points`; the picture is identical.
- Indexing a `Text` (`t[:6]`) counts glyphs without spaces. Count characters accordingly.
- Animations re-add their mobjects on top. Give markers and labels that must stay above
recolouring lines a `set_z_index(2)`.
- Text over lines needs `.add_background_rectangle(color="#0E1116", opacity=0.75, buff=0.04)`.
- Never stop a render with `pkill -f "manim ..."` from a shell whose own command line
contains that string; it kills the shell too.
- Each `-qh` render overwrites `media/videos/<file>/1080p60/<Scene>.mp4`. Copy a version you
want to keep before rendering again.
- Real-map recipe: project lon/lat with `x = (lon - lon0) * cos(lat0)`, `y = lat - lat0`,
build one polyline `VMobject` per route (downsample to ~70 points), then
`scale_to_fit_height` and `move_to` the whole group. Moving vehicles are `Dot`s on
`MoveAlongPath(dot, route, rate_func=linear)`.
Pliki w paczce (6)
- SKILL.md 7 KB
- fetch_clips.sh 1 KB
- narrated.py 6 KB
- review.sh 1 KB
- speed.sh 1 KB
- watch_render.sh 1 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · Alexzcoder/manim-skill | 100 | MIT |
| dreamlab-ai/agentbox | 100 | AGPL-3.0 |
| autonomous-ai/openharness | 100 | MIT |
| iart-ai/manim-skills | 100 | MIT |
| awesome-skills/manim-skill | 95 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 348f088b68b0 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100explainer-video
Tworzy animowane filmy wyjaśniające w stylu 3Blue1Brown: animacje Manim na czarnym tle, czcionka CMU Serif i narracja z syntezy mowy ElevenLabs, od tematu do sprawdzonego MP4 w 1080p60. - 96hermes-manim
Renderuje animację Manim na tym komputerze. - 100manim-video
Buduje wielokrotnego użytku wyjaśnienia animowane w Manim dla pojęć technicznych, wykresów, diagramów systemów i prezentacji produktów, jako czysty film wyjaśniający. - 100video-studio
Tworzy gotowy film od ujęć po scenę animowaną i lektor ElevenLabs, sprawdzając klatki i dźwięk przed oddaniem. - 100explainer-video
Tworzy krótki narrowany film wyjaśniający w stylu 3Blue1Brown na Macu: animacje Manim, lokalna synteza mowy Kokoro i kontrole dźwięku i etykiet przed udostępnieniem, dla kursu, wykładu lub własnego zrozumienia. - 100manim-explainer
Tworzy krótkie animacje wyjaśniające z matematyki i nauki w Python Manim: od koncepcji po scenę, polecenie renderu i sprawdzony plik.