Engines

Code-driven pipelines that when paired with an LLM, turn prompts into video. Each engine may produce a different output format, or artifact, and support different animation runtimes.

hyperframes

HyperFrames

Write HTML. Render video.

HyperFrames is an open-source framework by HeyGen that turns HTML, CSS, and JavaScript into deterministic MP4 videos. Compositions are standard web pages where timing and track data are defined via HTML data attributes. AI agents already write HTML, so they can compose, animate, and render videos without learning a proprietary format.

License Apache 2.0
Output MP4 (high resolution)
View details
28 models supported

Highlights

  • HTML-native: compositions are HTML files with data attributes, no proprietary timeline format
  • Adapter-based animation: GSAP, CSS, Lottie, Three.js, Anime.js, or WAAPI
  • Agent-friendly: non-interactive CLI designed for AI coding agents
  • Deterministic rendering: headless Chrome captures frames, FFmpeg produces output
  • Open source with no per-render fees or commercial-use thresholds

composition.html

<div data-composition-id="demo" data-width="1920" data-height="1080">
  <h1 class="clip" data-start="0" data-duration="5" data-track-index="0">
    Hello, HyperFrames!
  </h1>
</div>

lottie

Lottie

Vector animations, everywhere.

Lottie is a JSON-based animation format originally developed by Airbnb that renders high-quality vector animations on web, iOS, and Android. Instead of storing pixel data, Lottie files contain mathematical instructions for shapes, colors, and motion that a player library renders in real-time using the device's own graphics engine.

License MIT
Output JSON / .lottie (vector)
View details
28 models supported

Highlights

  • Vector-based: scales to any resolution without pixelation
  • Lightweight: JSON files are a fraction of the size of equivalent GIFs or MP4s
  • Cross-platform: native players for web (SVG/Canvas), iOS (Core Animation), and Android
  • Interactive: elements can respond to scroll, click, and hover events
  • dotLottie container format bundles multiple animations with themes and state machines

animation.json

{
  "v": "5.5.7",
  "fr": 30,
  "ip": 0,
  "op": 90,
  "w": 1920,
  "h": 1080,
  "layers": [...]
}

remotion

Remotion

React in, video out.

Remotion turns React components into MP4 videos with frame-accurate timing. Pipevideo returns editable Remotion source alongside the rendered video so you can review, tweak, and re-render.

License Remotion License
Output MP4 + Remotion source (TSX)
View details
28 models supported

Highlights

  • React + TypeScript: components, sequences, and hooks with deterministic output
  • Editable artifact: receive Remotion source alongside the rendered MP4
  • Provider-optimized rendering through Framerate
  • Same orchestration models as HyperFrames and Lottie

Composition.tsx

export const ProductTeaser = ({ title, accentColor }) => (
  <AbsoluteFill style={{ backgroundColor: "#000" }}>
    <Title text={title} color={accentColor} />
  </AbsoluteFill>
);

manim

Manim

Mathematical animation, rendered precisely.

Manim is the programmatic animation engine created by Grant Sanderson (3Blue1Brown) for mathematical and technical video. Scenes are Python classes with frame-accurate control over equations, geometry, graphs, and step-by-step derivations — ideal for STEM explainers where exactness and labeling matter.

License MIT
Output MP4 + Manim scene (Python)
View details
28 models supported

Highlights

  • Math-native primitives: LaTeX equations, coordinate systems, vector fields, and function plots
  • Python scenes: full programmatic control with deterministic, frame-accurate output
  • Step-by-step storytelling: transforms, reveals, and derivations animated precisely
  • Battle-tested by a decade of 3Blue1Brown-style educational video
  • Editable artifact: receive the Manim scene source alongside the rendered MP4

scene.py

class Pythagorean(Scene):
    def construct(self):
        eq = Tex(r"a^2 + b^2 = c^2")
        self.play(Write(eq))
        self.play(eq.animate.scale(1.4))
        self.wait()

mosvg

MoSVG

Vector scenes with verified motion.

MoSVG renders declarative motion-SVG: an SVG scene plus a JSON motion spec, rasterized deterministically to video by a normalize → synthesize → verify → render pipeline. Motion is verified with MoVer before rendering — direction, ordering, and rest states are checked instead of hoping free-form animation code behaves.

License Proprietary
Output MP4 / GIF + embeddable HTML
View details
28 models supported

Highlights

  • Declarative motion spec: per-object tweens with typed properties and easings, not free-form scripts
  • MoVer-verified motion: direction, ordering, and rest states checked before a frame renders
  • Single-rig scenes: logos, icons, and illustrations that move as one coherent vector scene
  • Deterministic rasterization: the same scene and spec always produce the same video
  • Editable artifact: receive the SVG scene and motion spec alongside the render

spec.json

{
  "fps": 30,
  "duration": 2.8,
  "objects": {
    "logo-mark": {
      "initial": { "opacity": 0, "sx": 0, "sy": 0 },
      "tweens": [
        { "p": "opacity", "t": 0, "d": 0.28, "to": 1, "e": "power1.out" },
        { "p": "sx", "t": 0, "d": 0.7, "to": 1, "e": "back.out(1.6)" }
      ]
    }
  }
}

All models work with every engine

Pick any orchestration model and pair it with any engine. Pipevideo handles the routing, authentication, and response normalization.

Browse all models
Video Generation Engines | Pipevideo | Pipevideo