Physarum Slime Mould

Four rules, and nothing that draws a network

Physarum polycephalum is a single-celled slime mould with no brain and no nervous system, and it is very good at finding efficient routes between food sources. Jeff Jones published a model of it in 2010 that reduces the whole behaviour to four rules applied to a population of simple agents, and it is that model — not an impression of it — that runs here.

Each agent holds a position and a heading, and nothing else. On every step it SENSES the trail at three probes, one straight ahead and one to either side; it TURNS toward whichever probe reads strongest; it MOVES forward; and it DEPOSITS a little trail where it lands. Then the trail map itself diffuses slightly and decays slightly, and the whole thing runs again.

That is the entire simulation. There is no code anywhere that draws a vein, a branch or a junction. The transport network is what those four rules produce when a quarter of a million agents run them at once, which is the point — the structure is grown, not illustrated. It is the same standard every generative effect here is held to: never fake the look with noise.

The image is the food map

In the real experiments you place oat flakes in a dish and the mould grows the network that connects them. Here your photograph is the nutrient field, and that is not a metaphor bolted on afterwards — the image's own structure is baked into the simulation state and read by every sense probe, for free, on every step.

FOOD chooses which part of the picture counts as nutrient. Bright feeds the agents on highlights, Dark on shadows, and Edges on the places where the image changes fastest, which tends to grow a network that traces contours and outlines rather than filling regions. None removes the food entirely and lets the colony organise on its own — worth doing once, because the difference between that and any of the other three is the clearest demonstration of how much the picture is actually steering things.

The food map does two jobs at once: it decides where agents are BORN, by importance-sampled spawning, and it decides where they keep heading. So a different photograph does not give you a recoloured version of the same network. It gives you a different network.

Why the dish is bigger than the picture

The agents live in a container and bounce off its walls, because wall-following is real slime-mould behaviour and taking it out makes the colony behave less like the organism. The first version of that drew a hard black rectangle around every render — the wall was inside the frame, and the trail piled up along it.

The fix was not to weaken the physics but to move the frame: the dish is four per cent larger than the picture on every side, so the wall and its build-up sit outside what you see. You get the behaviour without the border.

It can steer other effects

Physarum is the first effect in V3KTR that PRODUCES the shared field channel rather than consuming it. The trail is a scalar — a single strength per point — so what it publishes is the flow that trail implies: the direction running ALONG the veins.

Put a Physarum layer above Liquid, Displace or Point Cloud and set that effect's Field to FIELD CHANNEL, and the picture is moved along the network the mould just grew. Dye smears down the veins; points travel the branches. Neither effect had to know the other existed, and Physarum gains no control for it — it publishes only when something beneath is listening, and a Physarum layer with nothing under it is simply Physarum, showing its network.

What the controls are really doing

Agent count and Growth are the two that decide what you get. Agents sets how many organisms are competing for the same food, and Growth is how many simulation steps run before you see the result — a young colony is exploratory and sparse, an older one has pruned itself into a few strong routes. Watching the same settings at different Growth values is the fastest way to understand the model.

Deposit and Decay are in tension by design: deposit lays trail down, decay takes it away, and the ratio between them sets how strongly the colony reinforces a path it has already found. High reinforcement gives thick, decisive, few-route networks; weak reinforcement gives a fine exploratory mesh that never quite commits.

RENDER decides only how the finished trail is drawn, and changes nothing about the simulation — Ink and its inverse, Tinted, Image Ink, Etch and Glow. Image Ink is the one to try first if you want the photograph to stay present: the network carries the picture's own colour rather than replacing it.

When you'd reach for it

When you want structure that came out of the image rather than over it. A texture overlay looks the same on every photograph; a grown network does not, because it was grown from this one.

It is one of the heavier effects here, and deliberately bounded — the agent count and step count are capped so a simulation cannot run away from the GPU. Put it early in a stack and treat what it produces as new material for the effects beneath it, rather than as a finish.

Every option, from the app

Generated from the registry the app publishes, so this list is whatever the current build actually ships — not a copy of it.

Food

BrightDarkEdgesNone

Snap

Off90°60°45°30°

Render

InkInk (inv)TintedImage InkEtchGlow

Related effects

• Reaction–DiffusionA Gray–Scott reaction–diffusion simulation seeded by your image, with curated feed/kill presets for spots, stripes, worms, mitosis, coral and chaos.
• Point CloudEvery pixel lifted into real 3D space by inferred depth and rendered as billboarded splats with true depth-buffer occlusion — six displacement fields, four materials, three environments, and a field channel another effect can drive.
• MetaballDepth points scattered into a volumetric density field and raymarched as an isosurface — blobs that genuinely fuse where they meet, with capsule connectors, cast shadows and ambient occlusion.

Open the app to try it; the free effects work with no signup. Open V3KTR · all the FX · how it works