When I first heard that NFT collections like CryptoPunks or Bored Apes were “generated,” I assumed it meant low effort. A computer spits out random images, someone sells them — that was my mental model. It took me a while to understand that generative art has a longer history than NFTs, and that the algorithm is the creative work, not a shortcut around it.

Generative art is art created through a set of rules or an algorithm. The artist writes the system — the rules, the variables, the possible combinations — and the system produces the output. In the context of NFT collections, this typically means thousands of unique images produced by combining different traits according to defined probabilities.

How It Works in Practice

A generative NFT collection usually starts with a set of layered traits — backgrounds, bodies, accessories, expressions, colours. Each trait has a defined rarity. A gold background might appear in 1% of outputs. A common background might appear in 40%. The algorithm combines these layers randomly within those probabilities to produce each unique image.

The result is a collection of thousands of images that share a visual language but are each distinct. No two are identical. Some combinations are extremely rare, which is why certain NFTs within the same collection trade at much higher prices than the floor price. Rarity is built into the generation process from the start.

On Polygon, the low gas fees make it practical to mint large generative collections without the cost becoming a barrier. A 10,000-piece collection that would be expensive to deploy on Ethereum mainnet becomes much more accessible on Polygon — both for creators and for buyers minting individual pieces.

On-Chain vs Off-Chain Generation

There’s an important distinction in how generative NFT art is stored. Some collections store the actual image files off-chain — on servers or via IPFS — and only the token metadata lives on-chain. Others store the generative algorithm itself on-chain, meaning the image can be recreated from the blockchain data alone, without relying on any external server.

Fully on-chain generative art is considered more durable — if the platform disappears, the art can still be rendered from the blockchain. Off-chain storage depends on whoever is maintaining the server or IPFS pin continuing to do so. This distinction matters more than most people realise when they first buy into a collection.

The Connection to PFP Collections

Most PFP NFT collections are generative. The 10,000-item collections that became cultural touchstones during the NFT boom were almost all produced this way. The generative process is what makes each token unique enough to qualify as an individual ERC-721 token rather than copies of the same thing.

The royalty implications are also worth noting. When a rare generated piece sells on secondary marketplaces, the creator receives a percentage of the sale — if royalties are enforced. For popular generative collections with high secondary trading volume, those royalties can be significant.

What I’m Still Figuring Out

The line between generative art as a creative practice and generative NFTs as a speculative vehicle is blurry and I haven’t fully sorted it out. Some projects genuinely explore what algorithmic creativity can do. Others use the language of generative art to sell 10,000 variations of the same cartoon. I don’t always know which is which until I look more carefully at the project.

What I do think is that the underlying technology — algorithms producing verifiably unique on-chain assets — has more interesting applications ahead than profile pictures. That’s the part I’m watching.