AI Oracle - NFT Authenticity Verification
Early detect the on-chain NFT infringement to protect the creators & collectors from the fake NFT problem.
This AI Oracle is the first part of the NFT Audit services in the form of a Similar NFTs Detection report and can be integrated into any NFT Platform.
Using an AI Oracle by Oraichain - Similar NFT Detection AI Model as the core technology, our Content Authenticity Verification service allows users to upload an asset (Image, Audio, Video) and generate a report, which shows similar NFTs collected over many blockchains and marketplaces, protecting IP for creators and giving users more clues to evaluate the authenticity of the NFT before purchasing them. This AI-based mechanism will give both new and experienced NFT collectors an easy-to-use this engine to protect themselves from fraud.
The AI-based Authenticity Verification Service for NFTs is available on aiRight, activating the ability for NFT Collectors to easily review the on-chain data necessary to make an educated purchase decision for non-fungible assets.
In the form of a Similar NFT Detection report, this service will help users to view all similar NFTs found over supported networks and marketplaces to compare and have a more convenient way to investigate the relevant NFT. The information goes along with the founded NFT includes:
- NFT name and ID
- The direct URL to the NFT on the platform/marketplace
- Minting contract
- The creation time
There are two main core factors that contribute to the generation of a report, the first is building on-chain and off-chain databases, second is the AI Model scanning database and giving checking results.
We have built several databases as reference sets for authenticity evaluation. The on-chain NFT databases are collected from NFT marketplaces. The off-chain databases are collected from public sources related to artworks, music, and movies.
The database contains 5M+ NFTs from mostly Ethereum - the largest blockchain for NFT minting.
Our Database is being quickly expanded to the blockchains that NFT is popularising:
The match/non-match decision is challenging since there is not only one but there are several levels at which two media (two images, two audio files, one image and one video, etc.) can be considered “similar”. In the most restrictive case, the “exact duplicate” is detected bit-wise. In other less restrictive cases, slight changes in brightness, noise, or the addition of small perturbations are also considered “similar”.
In the most dynamic cases, two instances with the same object (e.g., same person) or different objects but with the same appearance (colour or rhythm) may be also considered “similar”. We, therefore, define several levels in which the two instances can be decided if they are copies of each other and arrange them with the following priority in the Similarity check report
There are several types of Near-exact duplicates and Edited copy levels:
- The Near-exact duplicate includes: Change of aspect ratio, Blurring, Brightness adjustment, Cropping, Additive noise, Padding, Rotation, Saturation adjustment, Jittering, Adding stripes Text overlaying
- The Edited copy includes: Change of aspect ratio, Blurring, Brightness, Noise, Perspective, Saturation
1. Exact duplicate
2. Near-exact duplicate (Light transformation) Near-exact duplicates correspond to light level of changes that are nearly human imperceptible such as compression distortion.
3. Detect a level of manipulation and variation: Edited copy or instance copy (Hard transformation) Edited copies correspond to an image pair where both images are edited version of the same source image with a high level of transformation.
- Here are some standard examples of particular type of transform with 2 level of transformation (light and hard transformation) for your better understanding about near-exact and edited copies:
Oraichain provides an open-source API of this AI Oracle to integrate into any NFT Platform.