Aria

Finished videos with real-sounding voices, made from code, by your coding agent.

Aria

Ask your agent for a video and Aria takes it the whole way, on your own machine:

script  →  voice  →  captions  →  scenes  →  render  →  check  →  share

Voice that whispers, laughs, gets excited, gets dramatic: inline emotion tags ([happy], [whispering],
[dramatic], [chuckle]...), cloned voice presets, and every take heard back by speech recognition and re-spoken
until every word is right. Word-level timings come out as captions.json.
Video written as React + TypeScript: karaoke captions, motion graphics, real 3D (lights, materials, 3D text,
bloom) and an original score in code, all cued to the exact spoken word, rendered on your GPU.
Checked before you see it: clipped text, safe areas, contrast, loudness and compression artefacts, then
encoded into a share-ready MP4.

The showcase film in examples/showcase was made entirely with Aria: narration, captions, 3D,
music and render. Watch it in the latest release.

Quick start

Needs Windows 10/11, a GPU with Vulkan or DX12, ffmpeg on PATH, and for voice an NVIDIA GPU (~6 GB VRAM).

In Claude Code:

/plugin marketplace add Abhinav-murugan1/Aria
/plugin install aria@aria

Then the local toolchain (render engine, voice environment, the aria command):

git clone https://github.com/Abhinav-murugan1/Aria; cd Aria
./install.ps1 -EngineOnly -Voice
aria doctor

Codex, Cursor, Gemini CLI, OpenCode, Windsurf, Cline, Kiro, Qoder and others: see INSTALL.md. Every
route needs the same local toolchain.

Then just ask:

make a 30 second vertical short explaining why we procrastinate, warm narrator, with captions

make a 16:9 launch video for my app, dramatic voice, 3D logo reveal, upbeat music

brag about this project: 20 seconds, cinematic

re-voice line 3 with more excitement and make the title land on the word "now"

command	does
`/aria`, `/aria-video`, `/aria-voice`	load the skills themselves
`/aria-help`	list every Aria command
`/aria-short TOPIC [9:16\|16:9\|1:1] [voice] [length]`	a finished video end to end (9:16 by default)
`/aria-brag [project dir\|URL] [--tone T] [--format F] [--duration S] [--no-voice]`	a 20 s launch video of your project or site: narration, music, poster, share copy
`/aria-new NAME [shape]`	an empty project from the template
`/aria-narrate SCRIPT.json [OUT_DIR]`	voice a script: emotion tags, hearing check, word-timed captions
`/aria-retake LINES [direction]`	re-speak lines, e.g. `/aria-retake 3,7 more excited`
`/aria-preset NAME KOKORO_VOICE\|REC.wav`	a new expressive voice preset (recordings need your consent)
`/aria-emotion WAV...`	the emotion meter, explained
`/aria-preview [EDIT.tsx]`	stills of every scene on one contact sheet, reviewed
`/aria-make [EDIT.tsx] [NAME]`	render, share copy, export check, fix (at most 3 versions)
`/aria-check VIDEO.mp4`	the export check on any video, with a fit-to-publish verdict
`/aria-doctor`	check the environment and say how to fix it

The command-line tool is lowercase aria (aria new, aria voice, aria make, ...); the product is Aria.

Only clone voices you own or have permission to use. Publishing is always your call.

Voice models: Chatterbox / Chatterbox Turbo (Resemble AI, MIT),
faster-whisper (MIT),
Kokoro (Apache-2.0). Music in code: Tone.js.
The aria-brag launch-video method is adapted from /brag by Shunit
Haviv Hakimi (MIT; see skills/aria-brag/NOTICE).