determined.ink

Cryptographic Proof-of-Work Generative Art

About

determined.ink is a generative art project exploring a new technique of creating images by mining them pixel by pixel. The output is shaped by a signal the process cannot control. What you see above is a piece being created in real-time.

Each pixel comes from a SHA-256 hash translated into a colour: three values for red, green, and blue. The miner tries hundreds of thousands of nonces to find a colour of a pixel that satisfies aesthetic rules set for that piece, for example how close it must stay to its neighbours. Those rules are the difficulty. The stricter they are, the more computation each pixel takes, and the more the look of the image is shaped by the constraints themselves.

An open educational extract of the foundational algorithm is on GitHub: github.com/mbrucki/pixel-mining.

The Manifest

The process exists not as a hidden means toward an image, but as the visible result of creative work. The process is not chosen for symbolic or aesthetic reasons. It is technically necessary. Without it, the image could not exist in its final form.

The piece is not rendered in advance, nor gradually revealed from a predetermined structure. After each pixel is mined, the system does not know what the next pixel will be. The only way forward is to continue the work. There is no shortcut, no closed form, no possibility of skipping the process without losing the result.

The process has a beginning, a duration, and an end. It is initiated by a seed, unfolds through mining, and resolves in a completed image. The beginning matters, because the seed makes the piece deterministic: the first pixel is the initial condition, and everything that follows is chained to it. The end matters, because without it the process would remain unresolved. But the duration matters equally, because the image can only emerge through time.

What becomes visible is a form of value production that cannot be compressed into seconds, as is common in digital creation. The time spent is not an artistic gesture added afterward, but an unavoidable cost of the form itself.

Collection

The aim was an algorithm that makes visually satisfying works: a series that holds together, while every piece still feels different, interesting, and complete on its own. That search began in April 2024. More than two years later, v10.9 is the first version that reliably delivers it. Below are the completed live 1080×1080 pieces, newest first.

The Journey

I am trying to find an algorithm that produces satisfying visual output through pixel-by-pixel mining, where a batch of runs on the same rules does not turn monotonous. The journey started in April 2024, and it has been an exciting trip so far.

5 April 2024

First mined image

The first piece ever mined. I found out that SHA-256 can be translated into RGB, and at the same time how hard it is to get a satisfying visual result from that translation alone.

23 April 2024

Early batches

A few weeks later I reached something that read as a pattern. Mining difficulty was not Bitcoin-style (counting leading zeros) but constraints the translated RGB values had to satisfy before a pixel could be accepted: chiefly, fitting their neighbours. Pixels were mined horizontally, left to right, yet the pattern still drifted diagonal.

28 April 2024

960 × 540

For the first time mining crossed one full day. The same script on a larger canvas showed that visual coherence broke down on later rows: the formula was not yet strong enough to hold the whole field together.

14 October 2025

Mining Algorithm v1

Only after a year (with a fair pause in between) came something like a breakthrough: I could control the pattern, stop the diagonal drift, and keep bands horizontal.

20 October 2025

Mining Algorithm v1

I added a curated shape in the centre of the canvas and mined the first piece whose look I could stand behind at the time. That was when I decided to build a website with a live mining section and move production from my computer to the cloud.

16 January 2026

Mining Algorithm v1

First 600×600 run to test scale. The pattern collapsed: the lower part of the canvas lost colour and coherence because the formula still was not right for bigger surfaces.

17 & 26 March 2026

Mining Algorithm v4

Visual output became stable and satisfying on larger, more compute-heavy canvases — including a second 600×600 run nine days later to confirm it was not a one-off.

1 April 2026

Mining Algorithm v4

The first HD piece at 1080×1080. Mining took about twenty-one hours. The dithered background that worked at smaller sizes vanished at this scale; the result felt flat and disappointing.

2 April 2026

Mining Algorithm v4

I changed how the miner handles large canvases: fewer checkpoint and progress-image writes (every five hundred pixels instead of constant disk and upload traffic). Mining moved to a GCP c2d-highcpu-16 VM where the hasher ran as a C extension on Intel SHA-NI. Together that cut wall time sharply — about four hours per piece instead of a full day. Two more 1080×1080 runs; visually they were still not good enough.

12 April 2026

Mining Algorithm v5 OLH

First experiment with colours taken from a photograph I had shot: the miner could only draw from that palette, using optimal learnable hashing against a reference image resized to the canvas.

25 May 2026

Mining Algorithm v6.1

I dropped the palette-from-photo idea and pushed on form instead: where else is pixel-by-pixel mining useful? I tried to re-mine real photographs, using the colour at each original coordinate as the target for the corresponding mined pixel.

26 & 27 May 2026

Mining Algorithm v7

The re-mining experiment disappointed, so I tried a new form. For the first time pixels were chosen in walk order, not strict left-to-right: each SHA-256 digest selects the next cell from an active frontier. Colour targets were tied to live space-weather readings written into the chain root. A 1080×1080 piece finished in about two and a half hours — far less wall time than the earlier HD row-major runs.

30 May 2026

Mining Algorithm v7.9

More pieces in the same walk-order family, rules tweaked slightly — still frontier growth, still cosmic-driven colour, still mined one pixel at a time.

8 June 2026

Mining Algorithm v8.3

A clear break and a new direction: forms I cannot predict in advance. For this piece the colour palette was curated; with a free palette the same technique produced much weaker results. Despite a large speed gain, mining took five days, nine hours, and forty-one minutes — the longest run in the project so far.

July 2026

Mining Algorithm v10.9

More than two years after the first mined image in April 2024, v10.9 is the first algorithm that reliably produces satisfying works: coherent as a series, yet each piece still distinct. The full series is in Collection above; three pieces here mark the milestone.