RepositoriesInvokard Studio

Agent skill + Python engine · MIT

Invokard Studio

Reels that look edited, made by your coding agent.

A skill that teaches Claude Code, Codex, Antigravity, Gemini CLI or Cursor how to produce a reel, and a Python engine that composes every frame: word-by-word captions with the key word highlighted, animated headlines and chips, charts, before/after wipes, cuts on the beat, end cards and a mixed soundtrack.

  • 9:16 · 16:9 · 1:1
  • Python 3.10+ and ffmpeg
  • Windows · macOS · Linux
  • MIT code
Ekilib, 45 s: one of the three delivered reels in the repository.

Examples

Three delivered reels

Each one was made with the engine, and the reel.py behind it ships inside the skill, so an agent reads it before writing its own.

The example media are shown with permission to demonstrate the software. They are not covered by the MIT licence.

How it works

From a script to a finished file

The agent writes reel.py

Between 60 and 150 lines that describe shots, layers, voice and effects. The skill is the method: what to generate and in what order, how to time it, what to check and what things cost.

Timing comes from audio

One voice file per sentence gives the engine the pauses, and captions and cuts derive from them. With a recorded narration or a song, align.py (faster-whisper) gives the words.

Preview before spending

The whole edit is reviewed on still images, as a contact sheet. Clips are generated last, only for the shots that need motion.

Render and verify

Pillow composes each frame and ffmpeg encodes H.264 with a ducked, normalised mix of about −14 LUFS. media.py verify checks the final file.

Install

Clone, install, ask for a reel

Install the engine and the skill

git clone https://github.com/Octonove/invokard-studio.git
cd invokard-studio
python -m venv .venv
python -m pip install -r requirements.txt
python install.py doctor
python install.py

Activate the virtual environment before the pip line (source .venv/bin/activate, or .venv\Scripts\activate on Windows). install.py puts the skill where Claude Code, Codex, Antigravity, Gemini CLI, Cursor and Windsurf look for it.

Your first reel in five minutes

python examples/first-reel/reel.py --preview
python examples/first-reel/reel.py

No account, no API key and no downloads: it draws its own images and synthesises its own voice and music.

You need

  • Python 3.10 or newer
  • ffmpeg on your PATH
  • A media provider (Magnific, ElevenLabs) only to generate new images, clips or voice; your own files need none

What it doesn't do

Where it stops

  • It doesn't publish to Instagram or TikTok, on purpose: it delivers the video and a cover, and you publish.
  • Rendering isn't instant. Frames are composed one by one, about 50 ms each, so a 45-second reel takes a minute or two.
  • Generating new images, clips or voice needs a provider account and credits. Rendering your own media needs neither.

The code, the issues and the history are on GitHub.

Star it if it helps; open an issue if it breaks.

View on GitHub