MetaMelange

This project allows users to combine any two of their ERC-721 tokens to generate, through a complex machine learning algorithm, an expressive new NFT that is a unique combination of the two parent tokens with a provable lineage. As an example, in our demo, we blend one of my favorite Fidenza NFTs and a World of Woman NFT. The result is an expressive artwork that combines the qualities of both and has provable lineages on chain to its parent. Our project is compatible with any image-based NFT tokens, and solves a few major challenges we see in the NFT ecosystem. First, is reducing the outsized concentration into the largest NFT projects. By allowing users to blend different NFTs, we create an incentive to broaden your portfolio beyond the mainstream names. In this demo project for the hackathon, we've specifically chosen to allow image uploads so users can experiment to their hearts content. In the production version, users will be limited to the NFTs in their portfolio, so a well diversified portfolio will allow users to access the most unique blends, and experiment to find truly innovative melanges. The second key feature, is that it lets users who might not have artistic or coding skills, to participate in the design and minting of NFTs to express their individuality. Blending for users is as simple as choosing two of your artworks and fine-tuning parameters through a few easy to use sliders. Finally, we aim to create even more excitement around owning a large NFT portfolio. Our own team has had a blast experimenting with different blending combinations and minting a wide variety of interesting melanges. Changing features sizes, locating better color palettes, and finding the best match that speaks to our personal aesthetic has become an addicting game for us and we're sure many people will enjoy turning their own creativity loose with this technology and generate amazing, expressive artwork of their own. Ultimately though, our hope is to incentivize the NFT ecosystem to be a truly decentralized one, one where it isn't a clamor for a few large projects, but a vibrant ecosystem where users are excited to explore the long-tail.

At a high level, our project allows a user to blend two NFTs to generate a third based on the high level content of one image, and the style of the other. In the minting process, we also assign the newly generated NFT attributes that prove its heritage on chain. Our project uses machine learning to blend the two NFT images, specifically, inceptionV3, a model first developed in 2015. In order to manage the intense compute resources needed for machine learning, we use Tensorflow.js to compute the results of the model on the client's machine. This allows us to create a relatively lightweight Dapp without requiring image uploads or powerful GPU or TPU based servers. Inceptionv3 was chosen to comply with the ETHGlobal rules of no prior work in advance of the hackathon and is an off the shelf, open source implementation. This model is relatively old, and has no NFT specific training. Despite this, it performs relatively well at blending NFTs given the constrained resources and time we've given it. The blending itself could be dramatically improved by adopting more recent ML models and training them on NFT images, a time consuming process unsuited to this hackathon. This project is predominantly coded in vanilla javascript and uses sponsor technology to directly interact with the blockchain. Specifically, we utilize Tatum to generate and register new NFT contracts and mint new NFTs. The actual NFT images and data are uploaded and stored on IPFS through Storage.NFT. In order to pay for the minting process and to connect to the wallet, we utilize wallet connect to sign and execute a transaction. (Currently 0 ETH for testing purposes). These NFTs are immediately available for a user to sell on Opensea as soon as the mint transaction is processed by the Ethereum network. Our team is relatively small with just one developer, but we were able to quickly bring together the technology platforms of several sponsors, deploy a sophisticated ML model, and the demo you see today, allows users to actually mint a real NFT on the blockchain. (In the demo, a new NFT is generated and minted specifically on the Rinkeby Test Network)