Pastel Network is a platform catered to digital artists and non-fungible token (NFT) collectors. Unlike current services which either practice curation absentia or simply let ill-intentioned users run amok and sell copied artworks or inappropriate content, Pastel has developed their own solution which features NSFW filters, duplicate protection algorithms, and guarantees the permanence of digital works on the blockchain. By implementing Google deep learning models and designing around a fault-tolerant data retention system for file storage, Pastel seeks to become the Louvre of the twenty-first century through the use of masternodes (MN) which function as archival checkpoints to create the first timeless and trustless network for virtual collectibles.
Truly Decentralized Hosting for Your Artwork
One of the biggest questions regarding digital artworks is who truly owns the file connected to the token metadata? On Ethereum, at least, users are lucky to have a trustworthy URI and IPFS solution that guarantees that their animated .GIF is stored on chain forever. More importantly, where to turn when on-chain storage costs are too high? In many ways blockchain already has great storage solutions such as Sia and Filecoin. Make no mistake — Pastel is not a project concerned with blockchain file storage, yet it can benefit in every way and supply even the most demanding users with enough storage space for high quality animations and audio. At this point, such solutions are their own competitive sector that ultimately benefits Pastel, but the largest goal is to keep digital artworks of all sizes accessible for the indefinite future.
Current NFT projects on Ethereum store TOKENURI data on the cloud — in many cases this is just an AWS hosting setup which may not be paid for next year, or alternatively it is an Interplanetary Hosted FileSystem (IPFS) directory that has some serious constraints. In the past there have been times when a specific IPFS directory can’t be reached. This is a known issue that appeared in 2018 because the network grew too big for the current systems to parse. In addition to this, the high amount of network data throughput needed is a real thorn for users on metered connections. Moreover, pointing to dynamic filetypes is a technical mess; although IPFS supports both static and dynamic files, there is no possibility for server-side logic. In effect this means that any interactive or self-learning artworks are rendered an impossibility using today’s technology on Ethereum. And while centralized cloud services like AWS are an attractive hosting method for short term curation projects, the goal of any distributed network should be the deployment of a sustainable and reliable trustless and truly decentralized database that outlives a mere market cycle.
This presents a painful conundrum for artists and collectors alike. Not only are some digital “installations” currently impossible on IPFS, artists and collectors still have to ponder the fate of prized tokenized artwork in the coming year. Will this curation service pay for AWS hosting in 2021? Do they need an engineer to work around an increasingly complex data structure puzzle? There is no guarantee this will be stored years from now.
Much like BitTorrent, Pastel Network distributes “shards” across the network among masternodes. The basic Masternode architecture used in Pastel is also heavily based on the Dash project, which is itself built on top of the Bitcoin architecture and code base. This means that there are dozens, hundreds, or even thousands of network participants that are incentivized to keep small parts of each individual artwork and “seed” it or share with the network. In traditional decentralized protocols like BitTorrent this also leads to the “rare chunk” problem where 95% of the necessary data packages are easily found yet the missing files are unaccounted for, rendering the entire torrent ultimately useless.
Luby Transform rules, or LT-codes, allow for the reconstruction of missing data chunks and solving the rare data problem by making all chunks in the LT scheme fungible, meaning any chunk is as good as another chunk and they are ultimately in an equally mixed and spoon-fed “soup” of parts. To explain it another way, the network is able to communicate among masternodes and find the exact missing information through consensus where each node agrees that the proposed “rebuilt” file is identical to original ticketed version supplied by the artist. Pastel Network has implemented LT-codes specifically to ensure that artworks will be stored in a distributed manner and ultimately reconstructible even in the event of a significant network disruption.
Self-healing or self-balancing is possible through the use of erasure codes even after a master node leaves the network, or multiple MNs leave the network and a LT chunk is apparently lost. Each LT chunk for an artwork is given a random seed which can actually be reconstructed by the artwork registration ticket on the Pastel chain. The highest-ranked MN can retrieve enough adjacent LT chunks to reconstruct the original file and use the random seed corresponding to the missing LT portion to calculate the identical LT chunk. We believe this is one of the most unique features to any artwork-focused blockchain. Although AWS servers and other cloud hosting solutions may fade, by distributing content hosting among the nodes on the Pastel Network there is ultimately a healthier, long-term incentive structure for being a good steward for on-chain data.
Duplicate Checks and NSFW
The primary method for checking submitted artwork against duplicate images or animations on the internet is the image fingerprint vector, which is a particularly discriminative and robust method. Unlike a file hash, which changes completely when any bit of a file is altered, the fingerprint vector is relatively stable: the fingerprint is identical even when the image undergoes a variety of common transformations, such as cropping, resizing, adjustment of color/contrast/brightness, common “filters” used in image editing software, etc. And even if the fingerprint does change, the fingerprints of two related images will have enough tell-tale statistical similarities that we can readily discern the relationship by careful (but automated) comparison of the fingerprints. If any of these existing fingerprints are deemed to be “too similar” to the candidate image fingerprint, then the candidate image is rejected as being a duplicate.
This system will never block original artworks that are truly unique. Once this image fingerprint vector has been independently computed by at least 3 of the MNs, any new registration attempt for an image whose fingerprint is sufficiently similar will be rejected as invalid by network participants for the next N blocks. The purpose of this mechanism is to protect artists against a dishonest MN owner who attempts to register the image themselves, thus taking credit away from the rightful creator.
The NSFW check is also relatively straightforward, using an open-source deep learning model developed by Yahoo called OpenNSFW. This model, implemented on Google’s popular TensorFlow neural network architecture and trained on millions of inappropriate images, as- signs a score between zero and one for every image provided to it as input. If the resulting score is above a certain threshold, the Masternode rejects the submission as being unacceptable. For instance, if the model outputs a result that is too high — say .98, or so — it’s extremely likely to contain inappropriate content and the Pastel filter will prompt an initial block on this image or video. After all, there are other platforms for such things!
Pastel Network includes several features benefiting artists and collectors alike. By building a fully functioning blockchain around the idea of transparent, automated curation rules and ensuring permanent digital provenance through the use of distributed chunks among all network masternodes, Pastel has pushed the envelope for the nascent NFT industry. It is our hope that with new technology there will be a new class of digital collectibles on the blockchain — this time independent of the Ethereum network and technical constraints of IPFS hosting.
Inspired by this article and want to learn more about Pastel Network? Wait no more, and join the Pastel Network community via the channels below!