Generate stunning, physically-accurate fluid animations in seconds. Built on deep learning research in computational fluid dynamics, our AI corrects numerical errors in real-time for unprecedented accuracy.
The first tool that combines real physics simulation with AI correction, making professional-quality fluid animations accessible to everyone.
Our simulation engine uses MPS/MPH (Moving Particle Hydrodynamics), the same class of particle methods used in ocean engineering, tsunami modeling, and aerospace research. Every animation is physically accurate.
Our deep learning model detects and corrects numerical energy dissipation in real-time. Trained on physics simulation data, it achieves 2.4x improvement in wave amplitude preservation over traditional methods.
Touch and interact with the simulation in real-time. Customize colors, physics parameters, and rendering styles. Export as high-quality video, GIF, or embeddable code.
FluidMotion is built on peer-reviewed research in computational fluid dynamics and deep learning, developed at the University of Tokyo. Our core technology solves a fundamental challenge: suppressing numerical wave amplitude attenuation in particle-based simulations.
Industry-standard computational fluid dynamics technique deployed in tsunami modeling, dam break analysis, offshore engineering, and aerospace.
A neural network trained via imitation learning predicts correction accelerations from 16-dimensional physics features per particle, achieving 2.4x improvement in long-term wave preservation.
Optimized MPS engine running directly in your browser with spatial hashing for O(N) neighbor search and GPU-accelerated rendering.
// AI Correction Engine
// Input: 16D physics features per particle
// - position, velocity, density, pressure
// - velocity history (5 timesteps)
// - neighborhood statistics
// Output: correction acceleration (2D)
applyMLCorrection() {
for (particle of fluidParticles) {
features = extractFeatures(particle)
correction = neuralNetwork(features)
particle.velocity += correction * dt
}
}
// Result: 43.5% wave height preserved
// vs 18.4% without correctionFrom free exploration to professional production.