Skip to content

onimite v0.1.0

Choose a tag to compare

@github-actions github-actions released this 17 Nov 22:40
· 3 commits to main since this release

A four-line Lua script is typed and onimite render video.lua writes video.mp4.

onimite makes videos with code. A video is a Lua script: scenes, shapes, text, images, SVG and Lottie, video clips, 3D, shaders, physics, music, sound effects and spoken lines. One small native binary renders it to MP4, plays it in a live preview that reloads on save, and checks it: frames, contact sheets, problems found by script line, loudness and beats. It is built so that a coding agent can make a video end to end and know it is right without watching it.

Every piece of the suite is native Odin, with no browser, no Node and no bundled FFmpeg, and every shader in it, ours and yours, now goes through one compiler, oksl.

Install

curl -fsSL https://defnotmid.com/install.sh | sh          # macOS (Apple Silicon), Linux (x86_64)
irm https://defnotmid.com/install.ps1 | iex               # Windows (PowerShell)

Then, in a clone of the repository:

onimite render examples/why_onimite/video.lua -o why_onimite.mp4
onimite preview examples/why_onimite/video.lua      # plays with sound, reloads on save
onimite inspect examples/why_onimite/video.lua      # what is wrong, by script line

onimite update moves you to later releases.

With an AI agent

  • Claude Code: /plugin marketplace add defnotmid/onimite, then /plugin install onimite@defnotmid. That installs the skill and the MCP server.
  • Claude Desktop: open onimite-macos-arm64.mcpb or onimite-windows-x86_64.mcpb from the assets below.
  • Codex: $skill-installer install https://github.com/defnotmid/onimite/tree/main/skills/onimite-video.
  • Any MCP client: the server is the command onimite mcp. It has typed tools, returns frames as images, and has a guide tool that returns the skill.

What's in it

  • Scripts: scenes with overlaps and transitions, keyframes with easings and springs, and Remotion-style helpers (sequence, series, loop, stagger, a beat clock). Data comes in with --props, --set or --batch.
  • 2D (ovg): gradients, masks, clip paths, the 16 blend modes and every SVG 1.1 filter. Text is shaped (ligatures, Arabic, Indic, bidi) with colour emoji built in. SVG and Lottie can be overridden per id and per layer.
  • Video editing: cuts, speed, picture and sound from .mp4, .mov, .webm, ProRes 4444 and VP9 alpha clips. Subtitles and word-level captions.
  • 3D (o3d): view3d{} in three.js's vocabulary, with physical camera lenses, shadows, PBR, glass, toon and ShaderMaterial. It loads glTF 2.0 (skins, morphs, animations), 3MF and OBJ, and draws 2D onto 3D surfaces.
  • Shaders (oksl): SkSL, Shadertoy and GLSL ES, and multipass programs mixing them. They compile to SPIR-V and Metal, or to a CPU program for --software.
  • Motion blur and physics: a camera shutter for 2D and 3D alike. Box2D and Box3D bodies are simulated once at load, and impacts are times you can hang sounds on.
  • Sound (odawg): music composed from data with song{}, whose sections and notes answer the picture, plus one-shots with sound{}. The mix has ducking and a master limiter. Onsets and beats come from any clip.
  • Voice (ottis): say() speaks a line with every word's time, and the captions follow from it. say_cue lands animation on spoken words.
  • Output: H.264 / H.265 through the system's codecs (VideoToolbox, Vulkan Video, Media Foundation), AV1 from oicut's own encoder in every build, or PNG frames.
  • Checks: inspect, frame, strip, onion, motion, snapshot tests, audio analyze, and --json with stable error codes on every command.

Built from

library version
ovg v0.1.0 2D vector canvas, GPU and CPU
o3d v0.1.0 3D, a GPU renderer and a software one drawing the same pixels
oksl v0.1.0 shader compiler: SPIR-V, MSL, CPU (AVX2 JIT)
ocurves v0.1.0 curve geometry, exact coverage, booleans
oglyph v0.1.0 fonts and shaping, tested against HarfBuzz
oicut v0.1.0 codecs and containers, no FFmpeg
odawg v0.1.0 music engine where everything is data
ottis v0.1.0 text to speech with word timings

Each is MIT and usable on its own.

Known limits

  • Platforms: Linux x86_64 (AVX2, glibc 2.34+), macOS Apple Silicon (signed and notarized), and Windows x86_64 (unsigned, best effort, tested only in CI). There is no Linux arm64 build yet.
  • --software speed: its shaders and filters run through oksl's AVX2 JIT on x86-64 Linux. On Apple Silicon and Windows they run through oksl's interpreter, which is slower. GPU rendering, the default, is the same everywhere.
  • H.264 / H.265: these need a codec on the machine. Releases ship none of their own; --codec av1 always works.
  • say(): its voice models are an optional download of about 850 MB on first use. ottis tests it on Linux, macOS and Windows; onimite's own Windows build runs it untested.
  • Not yet: WebM and alpha video output.
  • Stability: this is 0.x: the script API and the CLI can still change, and every change is in the release notes.
  • Benchmarks: BENCHMARK.md races onimite against every fframes example on each release, and it lists the races fframes wins.

Questions and videos you made go in Discussions. defnotmid.com · @fluxin on X