Kyle Stargarden

Oct 27, 2020

5 min read

N_Safety Forgery Detection Model Untitled NFT Hackathon Submission

There’s a tremendous amount of hype surrounding non-fungible tokens (NFTs) in the blockchain world today. NFTs are promising and present many new use cases, value adds and enhanced functionality for a plethora of different industries; from games and decentralized finance to art marketplaces and blockchain education. The buzz surrounding the opportunities with NFTs is warranted, but there is a fundamental problem. There is a fast growing impersonation, forgery and counterfeiting community just below the marketplaces. These thieves are eroding trust, fracturing governance and tribalizing the nascent NFT marketplace community.

The gig is pretty simple. Copy some known creator’s artwork, make a fake profile using the branding and content of the original creator, and then post a fake collectible NFT that looks and seems like the original content. Every day there’s a new story of some hapless crypto collector who gets scammed for 5+ ETH. The problem is embedded in a very base layer of the ecosystem; and it must be addressed.

Feels bad

Trustless systems are a big part of crypto and are embodied in projects such as RTC (Read the Contract) and ideals like “Code is law”. While I am inclined to see the benefit of moving into that direction; getting average users to step into and adopt cryptocurrency in it’s various formats, requires trusted intermediaries. We are not all crypto-natives after all.

There are two NFT marketplaces that are my absolute favorites and I suggest that anyone check them out. These are Rarible ( ) and OpenSea ( ); these markets are pioneering an entire new realm of crypto collectibles. These folks are the OGs of the NFT ecosystem. They lead the vanguard and innovate in a space where no one had gone before.

Bravely go

I am now going to pick on Rarible. Why? They have chosen to innovate so rapidly that it’s become provocative. Rarible invented something they call “marketplace liquidity mining” and launched a governance token for their platform on the 15th of July this year, 2020.

This event was unprecedented, well received and new users flooded into the Rarible marketplace. It’s difficult to make a living as an artist but the promise of earning delicious Compound-style governance tokens was irresistible. This massive surge of new interest and users had some unforeseen effects.

The Rarible market discord governance channel is awash with the dissenting throngs.

It would seem that the launch of their new governance token and the massive flux of new users had ripped a hole in the social fabric of the platform.

The Rarible team is overrun. The market’s verification system and report systems seemingly require some manual labor. When that’s coupled with unprecedented growth — scalability issues arise, and the system breaks down.

Do I believe that Rarible is turning a blind eye to this behavior? Hell no. They are but women (and men). It’s simply not possible to handle a vast seething mob of artists and art thieves manually.


Introducing the N_Safety deep learning computer vision model for detecting and flagging forged art content on NFT marketplaces.

Kudos if you clicked on pitch deck =-D

With this model, we are able to address many of the primary barriers to adoption in the NFT marketplace ecosystem, like establishing trust and artistry provenance.

N_Safety indexes and stores image data from verified sources into a database, then checks newly minted NFTs against to verify their uniqueness and authenticity. The system is robust against almost all types of NFT forgeries including, copy and pasted, copy and moved, scaling, rotations and color change.

Moving Forward: There is still plenty of room to increase safety and trust on NFT marketplaces and there is still challenge.

  • AI modeling and predictions techniques are often require computational resources which are not free. Its difficult to absorb the cost of predictions by a platform because this is a potential attack vector.
  • Marketplace verification procedures are still somewhat manual. This could be automated with sufficiently robust artificial intelligence models.
  • Matic might be the solution for providing micro transactions which can empower low cost predictions for a variety of safety and trust enhancing features.

Presented by Bitscrunch ( )and ( )

Github Repo :

Bitscrunch team:

Saravanan Jaichandaran —

Ajay Prashanth —

Vijay Pravin Maharajan —

Ashok Varadharajan — team:

Kyle Webster —— Creative Futurist