chods is an NFT collection that uses AI to transform static image NFTs into a different style. A user inserts a chod token, of which there are 10000, and a reference to another existing NFT that is owned by the same user, and generates a bunch of variants in a theme that is particular to the chod. The user can then bind one of the variant images to the chod so that it can be picked up by third parties e.g opensea, twitter.

At first the binding is not permanent, and the chod can be re-bound, but eventually the generator will shut down, and the metadata for the final bound images will be snapshot and permanently recorded on chain.

###

With the high level overview out the way, I want to talk about the generator. At first I wanted to provide a huge variety of distinct styles. The idea was that a visually diverse set of 10k AI pics curated by many people would result in the most interesting generated NFT collection to date, but it turned out to be aesthetically jarring and the lack of cohesion would make it extremely difficult to identify members of the collection on sight.

Another problem is that even though we've all been blown away by what AI can synthesize by now, it is still less than ideal for creating stylistic variants of input images while preserving semantics with very high fidelity. In particular, highly detailed and color rich images so often have the unmistakable markings of AI.

The solution to both problems (cohesion and poor semantics) was to limit the styles to those characteristic of the most primitive minds: children and early humans

###

Though the generator will eventually be available to all chods, I decided to open it up to miladies first, in part because I wanted to release it very quickly, and because it lets me work out any problems with the system in production on a smaller scale. The tokens are 0.02 ETH each, I have 1000 for myself, and I won't authorize more than 1000 free mints.

If you decide to buy any before the generator opens up, sit tight because I'm working on it now. More information, including about the various styles, will be delivered along with new functionality. The aim is now to perfect the generator, so that it is able to produce consistently good looking variants across many kinds of inputs, while retaining a cohesive look.