Project Palingenesis: A Game of Life That Learns to Remember

Palingenesis — from the Greek palin (again) and genesis (birth): rebirth, regeneration, the recurrence of a thing.

This is a generative art series that treats a painting not as a fixed image but as a living probability landscape: something that decays, forgets, and tries to remember itself, one frame at a time. Under the metaphor there is a single engine, a heavily modified Conway’s Game of Life. This post is about that engine: how it regenerates, what parameters shape it, and the idea it is built around, which is attention.

I’ll use the three works currently public in the series:

• Lady with an Ermine (Leonardo da Vinci, c. 1489–1491)
• Portrait of a Young Woman (Sandro Botticelli, c. 1480–1485)
• The Birth of Venus (Sandro Botticelli, c. 1484–1486)

The Game of Life, briefly

You already know the rules. A grid of cells, each alive or dead. Every cell reads its eight neighbors and, at the same time as all the others, updates:

CELL STATE   LIVING NEIGHBORS   →   RESULT
──────────────────────────────────────────────────
alive        fewer than 2       →   dies (isolation)
alive        more than 3        →   dies (overcrowding)
alive        2 or 3             →   survives
dead         exactly 3          →   born

Four lines of logic produce gliders, oscillators, and chaos. No cell knows what it is building; the pattern is emergent.

That is the starting point. Everything else in Palingenesis comes from modifying those rules.

A Game of Life that regenerates

Classic Life is closed and deterministic. It runs toward emptiness or a repeating loop, and the same seed always produces the same result. It cannot remember anything, and it cannot surprise you. Palingenesis reopens the system in a few ways.

It is seeded from a painting. Before the first generation, every pixel of the source image becomes a cell. The probability that a cell starts alive is its brightness:

brightness = 0.299·R + 0.587·G + 0.114·B
P(alive)   = brightness / 255

Bright regions start dense, shadows start sparse. The painting is not drawn onto the grid; it is the grid’s initial state. From that moment the automaton starts losing it.

Death is negotiable. In classic Life a cell with the wrong neighbor count simply dies. Here a doomed cell gets a second roll to survive anyway. That chance depends on two things:

• a vitality parameter (meanRate), how strongly the system resists decay, and
• a wave field, two sine/cosine fields moving across rows and columns over time:

xWave = sin(x·0.02 + t) + cos(x·0.05 − t·0.5)
yWave = sin(y·0.03 + t·0.7) + cos(y·0.01 + t·0.2)
protection ∝ (xWave + yWave)² · vitality

So entropy is not uniform. It moves. Some regions of the canvas are, for a moment, more forgiving than others, and that boundary shifts over time.

Life can start from nothing. Dead cells in bright areas have a small constant chance of spontaneous rebirth, again weighted by the original brightness. The image keeps trying to reassert itself even where no living neighbor exists.

The system breathes. With autoPulse on, vitality oscillates on a slow sine wave between pulseMin and pulseMax (about 0.03 to 0.2 in these works). The canvas moves toward order and back toward dissolution on a regular cycle.

Color drifts, and the dead leave ghosts. When a cell is born it inherits its color from a random living neighbor, so palettes move and mutate across the population and drift away from the original hues. And a cell that dies does not disappear at once: a separate trail buffer fades it out by multiplication each frame (trail *= 0.92), rendered as alpha. What you see is never only the living cells but also a layer of recent afterimages.

Classic Life settles into order or emptiness. Palingenesis is built so it reaches neither. Decay and reconstruction run against each other continuously, and the image never fully resolves or fully disappears.

The main parameters, if you want to reproduce the behavior:

PARAMETER                        CONTROLS
──────────────────────────────────────────────────────────────
meanRate                         baseline vitality / resistance to decay
autoPulse, pulseMin, pulseMax    the slow oscillation of entropy
wave field                       spatial + temporal rhythm of protection
stochastic rebirth               spontaneous generation in bright regions
useDynamicColors                 color inheritance and drift
trail decay                      afterimage fade

Attention

Everything above describes decay. On its own it would erase the painting. To counter it, each work has a moving region where the normal rules are suspended and the original image is forced back to life: cells set alive and restored to their true colors. Where the region is, the painting reappears intact. Where it is not, entropy continues. In physics, a local agent that pushes back against entropy on its own patch is called a Maxwell’s demon, and that is what this region is.

It is also the same word used for transformer models: attention is where the system looks, and where it looks, the image holds.

Each of the three paintings gives that attention a different physical form.

Lady with an Ermine: attention as a reading head

Here attention is a triangle with three moving vertices. Every so often one vertex picks a new target and the shape glides toward it (lerp), always kept above a minimum area and locked to its winding so it never collapses. Any cell inside the triangle is restored to Leonardo's original color.

It works like a scanner, or an eye moving across a face. It wanders on its own and you do not steer it, so the portrait is never whole. It is a partial reconstruction shaped by where the triangle has recently been, closer to recalling a face from memory than to seeing it: some features sharp, others gone.

Portrait of a Young Woman: attention as water

For Botticelli’s portrait, the canvas becomes a pane of glass and attention falls as drops of water. Drops spawn at the top, descend with a slight wobble and jitter, and wherever they wet the surface the original image is restored. Then the glass dries: over a fixed number of frames the wet region evaporates and the automaton takes it back.

Here attention is temporary. It does not settle like the triangle; it rains down, briefly reveals, and evaporates. The image is restored along wet paths and lost as they dry.

The Birth of Venus: attention from within

The third work inverts the geometry. There is no external scanner and no falling water. Instead a mask confines the whole living system to Venus’s figure. The Game of Life (decay, rebirth, breathing, color drift) runs only inside her; everything outside stays as Botticelli left it.

Attention does not visit her from outside. She is where it happens. The life and the forgetting occur inside her form, flickering and dissolving within the silhouette while the rest of the scene stays still. The edge of her body is the edge of the whole process.

Why it matters to me

The project is really about one thing: a rule that operates at the scale of a single pixel, one cell counting its eight neighbors, producing at the scale of the whole canvas an effect of death and rebirth. Nothing in the code draws the painting dissolving or reassembling. That behavior emerges from millions of small decisions, many times a second.

Doing this to a five-hundred-year-old painting has two effects.

First, it makes a static object move. A painting we have stopped really looking at becomes something that changes while you watch and never repeats. The same source and the same parameters never produce the same run twice, because of the random survival rolls, spontaneous births, and drifting color.

Second, and this is the part I care about most, the constant dissolving and reforming moves the viewer’s attention. When the whole image cannot be present at once, the eye stops reading it as a familiar icon and starts noticing detail: the curve of a hand, the weave of a fabric, the light at the edge of a face. All of it was in the original, but it is the kind of thing you usually skip.

Palingenesis is a microscopic rule used to make a painting move, so that in watching it come apart and return, you notice what you had stopped seeing.

Palingenesis is an ongoing generative series about Renaissance paintings and the behavior of algorithms. More iterations to come.

I post Palingenesis pieces on X and on fxhash.