a world of evolving friends
This is a real-time evolution simulation. Dozens of tiny creatures live in a 2400×1800 pixel world divided into biomes. They eat, reproduce, and die — and over generations, their neural network brains evolve through natural selection and mutation. All behavior is learned, not programmed. No creature is told what food is, what a predator is, or how to survive. They figure it out — or go extinct.
Each creature has a neural network brain with 44 inputs, 24 hidden neurons, and 8 outputs (1,280 learned parameters). The brain controls everything: where to move, how fast, when to bite, when to reproduce, and what signals to broadcast.
Crucially, creatures have no hardcoded instincts. They don't know what food is, what other creatures are, or what danger looks like. They perceive the 5 nearest entities as raw sensory data — direction, distance, size, and color — and must evolve to make sense of it all. Over generations, creatures that stumble into useful behaviors (like moving toward food-colored things and biting) survive and pass on their brain weights.
Every visual trait is gene-driven and heritable:
A creature's diet is controlled by its dietType gene (0–1):
Creatures with high bite strength can attack other creatures — but only if their neural network learns to do so. Attacking costs 4 energy. Whether carnivores learn to hunt, and who they target, is entirely emergent behavior.
Each biome grows its own food types. Soft foods (berries, clover, flowers) can be eaten by anyone. Harder foods require stronger jaws:
Creatures start with 160–200 energy. They constantly burn energy just by existing (0.003/tick), plus extra for moving fast and being large. Crowded areas and overpopulation drain even more energy — the ecosystem has a natural carrying capacity. When energy hits zero, the creature dies — you'll see them wobble, squish flat with a cute splat, then fade away, leaving behind meat.
Snow slows creatures down by up to 20% and reduces food growth. Rain boosts food regrowth in most biomes.
Reproduction isn't automatic — a creature's neural network must decide to reproduce (repro_intent output > 0.5), and it must have energy above 55. Social creatures (sociability gene > 0.4) look for a nearby mate and combine genes — each gene has a 50/50 chance of coming from either parent. Less social creatures reproduce asexually by cloning their genes.
Either way, each gene has an 8% chance of mutating by up to ±0.3. The neural network brain is also cloned and mutated. The parent spends about 24 energy; the child starts with about 31. There's a cooldown of 2–3 seconds between births.
The world is divided into six biomes using layered noise patterns:
Those are germs — they appear when the population exceeds 150 as a natural population control. Germs drain energy from their host and can spread to nearby creatures. A creature's curiosity gene doubles as its immune system — higher curiosity means better chance of fighting off infection.
Both genes and brains. Over generations, natural selection shapes two things simultaneously:
Because creatures start with random brains, early generations are chaotic — they wander aimlessly and most starve. But creatures that accidentally move toward food and bite it survive longer, reproduce more, and pass on their brain weights. Common emergent outcomes include:
The Generation counter in the stats panel shows how many reproduction cycles the oldest lineage has gone through.
Each creature's neural network has three layers: 44 inputs → 24 hidden neurons → 8 outputs.
Inputs (what they sense):
Outputs (what they decide):
Mass extinction events can happen! Common causes: creatures failing to learn to eat in early generations, food depletion from overpopulation, a germ epidemic above 150 creatures, or harsh weather reducing food supply.
An extinction safety net prevents total die-offs: if population drops below 10 (early generations) or below 5 (later generations), the best survivors are cloned with slight mutations — preserving evolutionary progress rather than starting over. When population is critically low, creatures also burn less energy, giving them more time to find food and recover.
Tiny Creatures is a generative art & science toy.
No two simulations are alike. Enjoy watching life find a way.