A browser-native workbench for making generated worlds live.
Kaminos brings generated beings, live materials, local inference, motion, and spatial assets into one inspectable WebGPU workbench. It is where difficult technical systems become visible enough to steer, compose, and decide around while they are still alive.
The project spans four connected capabilities:
- Generated beings. Generated creatures can preserve deliberate morphology through generative transformation and return to mechanical control.
- Live materials. Simulated structure, appearance, and motion remain authorable while state advances.
- Browser-native intelligence. Spatial models execute inside the same local environment that inspects and consumes their outputs.
- A world kiln. Images, meshes, splats, motion, material fields, simulation state, and generated environments become composable world matter.
A live browser-native orange flame with a luminous crown and rooted filament structure
These films were captured directly from the live browser runtime while one stateful WebGPU combustion material was being authored, not prerendered.
The material carries its history through control changes. Existing momentum continues through contraction, acceleration, chromatic transition, changing source geometry, and renewed expansion. A broad burner can gather into a jet, retain the structure already in flight, and rebuild into another morphology without resetting the simulation.
Live Combustion presents one complete composition, four authored transitions, and seven compact studies of color, structure, width, and state history.
Kaminos currently carries a multi-field volumetric fire and smoke simulation that runs directly in the browser through WebGPU. Its current boundary-fire renderer derives compact structural fields around the combustion front, then uses those fields to guide where the volumetric renderer spends work.
The result is a live material process with:
- dynamic fire and smoke at interactive cadence;
- live simulation routes developed and profiled on Apple Silicon;
- combustion-front topology and baked boundary-sidecar fields;
- support, coverage, ridge, proximity, and footprint guidance;
- adaptive raymarch and explicit quality controls;
- route, backend, effective-control, and performance receipts;
- durable state capture and replay for visual investigation.
The fire began as an answer to a product question: how can a local AI experience remain alive while expensive inference occupies the machine? It is now becoming a material and rendering research program of its own.
Generated creatures can preserve deliberate morphology through generative transformation and return to mechanical control.
Deliberate edits to a parameterized creature template have produced corresponding changes after image generation and image-to-3D reconstruction. Using recovered correspondence, one reconstruction was registered to a control rig and manually skinned for large articulated deformations; another was driven by synthesized terrain-following motion.
That result reaches recovered structure and downstream mechanical consumption. The continuing research frontier is editable authority: returning distinctions authored by the generator as durable, named controls that can survive another generation and accumulate into later rounds of authorship.
Spatial models execute inside the same browser environment that consumes their outputs. Local WebGPU inference, geometry, motion, and simulation can therefore remain part of one operating world rather than terminating at a model response or crossing into a disconnected application.
The workbench preserves requested and effective route identity, model and backend lifecycle, generated assets, live state, and the inspection surfaces needed to decide what should happen next.
Kaminos treats generated outputs as world matter with lineage and behavior. Images, meshes, splats, motion, material fields, simulation state, and route outputs enter a shared browser workbench where they can be inspected, conditioned, transformed, staged, and handed onward.
Current substrate includes:
- Three.js/WebGPU scene editing and persistence;
- source-aware splat import, correction, crop, orientation, and sidecars;
- mesh/splat hybrid rendering and scene-context integration;
- motion-generation and motion-transposition experiments;
- browser-native fluid, particle, and material processes;
- world/chamber contracts for generated environments and inhabitants;
- route receipts that distinguish requested and effective execution;
- smoke and witness harnesses for visual and technical evidence.
The architecture is documented in Spatial Asset Kiln. Splat lineage and correction contracts are documented in Splat Assets.
The product direction is a complete visible loop:
source -> route -> live work -> generated matter -> inspectable cast
A team should be able to choose a source, fire a real local route, remain inside a living workroom while compute runs, and inspect the resulting object without crossing into a disconnected tool or dead waiting screen.
This loop is being assembled from the same browser-native systems already used for volumetric fire, splats, generated assets, motion, route scheduling, and scene inspection.
Kaminos projects are organized around one expensive uncertainty. The goal is a live artifact that makes the next product decision obvious enough to fund, staff, transfer, continue, or kill.
The recurring engineering move is to find the missing representation that makes the system tractable. In the current fire route, combustion-front and boundary fields turn an expensive volumetric search into guided rendering. In generated-world routes, the corresponding representation may be a depth or normal field, splat, mesh, semantic substrate, motion packet, route receipt, or world-state contract.
The visible experience carries the idea. The accompanying implementation, tests, route identity, performance evidence, and handoff notes make it usable after the demonstration ends.
Kaminos is developed as a local browser application. Serve the checkout:
python3 serve.py 8095
Open the current volume route in a WebGPU-capable Chromium browser:
http://127.0.0.1:8095/?kaminos_volume_smoke=1&volume_scene=tall_plume
The strongest current route is developed and witnessed on Apple Silicon. Other WebGPU devices may expose different performance and feature boundaries.
Kaminos is an active research and product prototype. Internal contracts, control surfaces, and route integrations are evolving quickly. Publicly useful subsystems are being separated into focused packages and upstream contributions as their interfaces settle.
For technical or collaboration inquiries, contact Noah Lyons.