hoshi

Web2 social media apps are flawed.

Today’s social media platforms — TikTok, Instagram, YouTube — are powered by creators. From viral dances to popular memes, users are constantly sharing new ideas. Yet, there’s a major problem: the original creators often go unrecognized and uncompensated. While larger creators and influencers who build on original trends or content reap the majority of credit and financial rewards, the originators of these ideas are not fairly rewarded.

The root of this problem lies in the inability of current Web2 platforms to effectively prove content ownership or enforce fair revenue sharing. Creators are forced into an 'all or nothing' revenue system due to rigid copyright rules. We need a solution to provide a fair, transparent, and equitable content creation ecosystem where intellectual property rights are protected and content contributors are compensated based on their value added.

hoshi aims to address these inequities by leveraging Web3 technology to create a platform where content creators can prove ownership, enforce fair use, and receive equitable compensation for their work. By giving IP rights to content creation through NFTs and an automated royalty distribution system, hoshi ensures that both original creators and those who contribute to or remix their work are fairly rewarded.

1. Key Features

1.1 Automated AI-Powered Detection Uses AI/ML to analyze content & determine an originality score across all other posts. We use an AI-based originality scoring system that quantifies the degree of contribution for each post. The more original the content, the higher the share of rewards allocated to the original creator. Users who remix or modify content receive proportional compensation based on their added value. An IP derivative tree is built based on these scores, to determine revenue flow sharing across all contributors based on their added value.

1.2 IP Ownership and Protection via NFTs hoshi ensures that creators retain full control over their IP by minting each piece of content uploaded to the platform as a non-fungible token (NFT). This provides verifiable proof of ownership and a permanent, immutable record on the blockchain, safeguarding against unauthorized use or plagiarism.

1.3 Smart Royalty Distribution hoshi employs a smart royalty distribution model that rewards original creators and those who contribute to or build upon their work. This system ensures that all contributors along the creative chain are fairly rewarded. Royalties are distributed through our native ERC20 token backed by ETH. When a user interacts with a piece of content, some of their tokens are transferred from their wallet to the wallet of the content creator, and through our smart contract code, royalty distributions of derivative work is automatically paid.

1. Sustainable Tokenomics Model hoshi's business model revolves around companies depositing ETH into our smart contract for advertising rights as they would for advertising rights with Web2 social media like YouTube and TikTok. However, instead of the centralized authorities taking some of the funds to build their company, 100% of the deposited funds go back to the users of the platform, in the form of our native token - hoshi tokens. For every ETH deposited, 1 million new hoshi tokens are minted in the liquidity pool in the smart contract. These tokens are then distributed equally across all users of the platform and are also used to pay for gas fees for any transactions that occur on the platform. Essentially, it is completely free for users of the app!

2.1 Subscription-Based Revenue Withdrawal Creators can directly withdraw the underlying ETH tagged to the hoshi tokens and directly earn funds from their content. This is done by enabling a whitelist of wallet addresses who are allowed to unwrap the tokens, so as to prevent casual viewers from unwrapping their tokens as well. To join the whitelist and make money off your content, hoshi adopts an approach similar to twitter that requires content creators to pay a subscription fee to earn money off their content on our platform. hoshi reinvests the subscription fees back into the platform’s liquidity pool. This subscription model not only facilitates monetization but also ensures the platform’s long-term financial sustainability.

1. Ecosystem Growth As hoshi grows, it creates a self-reinforcing cycle where increasing content creation leads to more user engagement, which in turn attracts advertisers and generates higher revenues. The platform’s transparent and fair revenue-sharing model incentivizes quality content creation while encouraging users to actively participate in the ecosystem. Over time, this virtuous cycle will lead to the continuous growth of both the hoshi token economy and the platform’s user base.

We believe hoshi is the future of social media content creation, and we hope that you will join us in empowering creators with the recognition and rewards they deserve.

To detect Copyright Infringement, we develop a proprietary AI pipeline to not only find the media and text that are similar in nature to the new post but also detect the specific location with the highest similarity.

To detect textual similarity of captions, we use Cohere Embeddings to embed each caption into a vector. When a new post is provided, we similarly embed the new caption and compare against other vectors in the database. However, such an approach heavily focuses on semantic similarity since copyright infringement requires exact textual lifting, we utilise the BM25 method to conduct full text search as well. The textual similarity score is thus a linear combination of the results from the vector and the full text search, making it robust and efficient.

For the media detection, we use a similar approach using SIGLIP (https://arxiv.org/abs/2303.15343) embeddings and compare the input media using these embeddings. Again following the need for similarity beyond semantic meaning, we apply structural similarity scores based off the histogram of the image. We compute the following matrix (Correlation of Histogram, Intersection of Histogram and the Structural Similarity Index (https://www.imatest.com/docs/ssim/).

To determine the location of the image that is most likely to be an infringement, we utilise a novel image localisation approach. We split the new image into many small patches and run a sliding window over the patches. Each patch in the window would have the same similarity score as all other patches in that particular iteration. After normalizing the scores, we would have a distribution of patches with varying similarity scores corresponding to a heatmap of the mini patches that are most similar to the existing original image.

To extend this approach to videos, we split the video up into frames at a specified frames per second and perform the search over each frame. The final original result is determined by majority voting which makes our approach temporally robust.

We made use of Dynamic Protocol for the user authentication. Making it easier for people to onboard onto Hoshi by allowing for email and Google account login and hiding login via wallets.

We made use of ERC20 and ERC721 token contracts by openzeppelin so as to create custom logic to our tokens and NFTs to fit our needs. The metadata of the posts/NFTs are uploaded to IPFS, further adding to the decentralization of our platform.

For the frontend, we used NextJS for the framework. We also used Three.js to create the 3D derivative graphs to show the connections between an original post and its derivatives.