relation
Cosine similarity becomes pairwise force. Relationships stronger than the field average attract; weaker relationships repel.
signal / live study
SEMANTICPHYSICS
What if meaning were a physical force?
Choose three concepts. A small local embedding model measures how their meanings relate, then the browser translates those relationships into attraction, repulsion, mass, orbit, and turbulence. The model never generates the picture. It changes the laws the picture obeys.
drag through the field to disturb it · arrow keys move the probe · space adds an impulse
procedural preview ready
strongest relation—
field tension00%
semantic spread—
runtimeprocedural
Semantic inference is optional and local. Activating it downloads an open embedding model to this browser. Your three concepts are embedded on-device and are not sent to a Warp the Surface inference service.
under the surface
The words are not prompts. They are constants.
mass
A concept that sits closer to the other concepts gains semantic centrality and therefore more gravitational influence.
turbulence
Concepts farther from the shared semantic center inject more instability into the surrounding particle field.
disturbance
Your touch does not rewrite meaning. It temporarily bends the resulting system and lets the semantic structure recover.
system
Intelligence determines relationships. Simulation determines form.
- simulation
- Canvas 2D force field + particles
- semantics
- mxbai-embed-xsmall-v1 / 24.1M parameters
- runtime
- Transformers.js worker / WebGPU preferred, WASM fallback
- model license
- Apache-2.0
- server inference
- none
- status
- experimental / mobile-first v0
The semantic model runs only when concepts change. Every animation frame after that is ordinary browser simulation. This keeps intelligence sparse and interaction continuous: the model supplies relationships, not rendered content.