DyNCA: Real-Time Dynamic Texture Synthesis Using Neural Cellular Automata

Figure 1. Our DyNCA model can synthesize infinitely-long realistic dynamic texture videos with arbitrary size in real time. Target Appearance: DyNCA learns a desired texture pattern from a given target appearance image. Target Dynamics: DyNCA can learn motion from different target sources. We allow the users to define the desired motion either by a hand-crafted optical-flow image 1 or a dynamic texture video. Synthesized Result: DyNCA synthesizes realistic dynamic texture videos. Each synthesized video frame resembles the target appearance, while the concatenation of frames induces the motion of the target dynamics. See our real-time interactive demo at 2 .