Kira Engine
The framework that powers real-time video transformation. Run any AI model. Transform every frame. 60fps, every device.
Two transformation paradigms: programmable effects (you control every pixel) and AI models (stream-to-stream transformation). Use one or combine both.
A New Paradigm for Video
Kira Engine gives you two ways to transform video: write code that runs every frame with access to detection data, or plug in AI models that transform the entire stream. Combine both for infinite possibilities.
Programmable Effects + AI Models
Choose your approach or combine both for maximum creative power.
Programmable Effects
Write code that runs every frame. Full access to detection data (face, pose, hands) and the video stream. Draw anything, anywhere. You control every pixel.
- Frame-by-frame code execution
- Access to 478 face landmarks
- Pose and hand tracking data
- Full canvas/WebGL access
AI Stream Transformers
Plug in AI models that take your video stream and output a transformed stream. Style transfer, face swap, diffusion - the model handles everything.
- Stream-to-stream transformation
- Style transfer in real-time
- Face swap and morphing
- Any diffusion/generative model
Combine Both
Use detection data to control AI models, or layer programmable effects on top of AI-transformed streams. The pipeline supports any combination.
Example: Use face detection to trigger style changes, then overlay custom graphics on the transformed output.
Building Blocks, Not Black Boxes
Kira Engine is modular by design. Assemble only the components you need. Extend with your own.
Plugin System
Everything is a plugin. Detection models, transformers, renderers - mix and match to build exactly what you need.
- Modular architecture
- Hot-swappable components
- Custom plugin API
Opt-in Loading
Nothing loads by default. Activate only what you need - no bloat, maximum performance.
- Zero overhead by default
- Lazy loading built-in
- Bundle what you use
Bring Your Own Model
Load any AI model. ONNX, TensorFlow.js, or custom models. The pipeline handles inference and frame timing.
- ONNX Runtime Web
- TensorFlow.js compatible
- Custom model support
Built-in Detection Layer
Battle-tested detection plugins powered by MediaPipe. Get tracking data for your effects - all optional, all opt-in.
All plugins are opt-in. Only loaded when explicitly enabled.
Face Detection
Complete face analysis with landmarks, blendshapes, and transformation matrices.
- 478 facial landmarks
- 52 blendshapes for expressions
- Face transformation matrix
- Multi-face support (up to 10)
Hand Detection
Precise hand and finger tracking for gesture-based interactions.
- 21 landmarks per hand
- Both hands simultaneously
- Finger joint positions
Pose Detection
Full-body skeleton tracking for immersive avatar experiences.
- 33 body landmarks
- Real-time skeleton data
- Upper and lower body
More detection plugins coming - segmentation, depth estimation, object detection.
Engineered for Real-Time
Video transformation demands zero compromises. Every component optimized for smooth, consistent frames.
Locked 60fps
Intelligent scheduling and parallel execution. Smooth output regardless of complexity.
Smart Memory
Automatic resource pooling, lazy loading, GC tuned for continuous video.
Universal
Adaptive quality scaling. Desktop, laptop, tablet, mobile - same code, every device.
GPU Accelerated
WebGL 2.0 compositing with hardware-accelerated rendering.
Your Choice
Run locally or use cloud models. You decide where processing happens.
Pure Web
No plugins, no installs. If it runs a browser, it runs Kira Engine.
What Will You Build?
The infrastructure for the next generation of video experiences.
VTubing
Full-body avatar puppeteering with expression sync and gesture control.
Face Transformation
Real-time face swapping, morphing, and identity transformation.
AR Effects
Interactive filters, face paint, virtual makeup, augmented overlays.
Style Transfer
Apply artistic styles, cartoon effects, or custom aesthetics to live video.
The Future of Video is Transformable
We're building the infrastructure for AI-powered video. For developers, creators, and companies pushing boundaries.
Building a video app? Training custom models? Creating next-gen experiences? Let's build together.