Description
Prompt marketplaces are crucial in the AI landscape, meeting the demand for skilled prompt engineers while democratizing prompt creation. These platforms provide affordable access to prompts, enabling hobbyists to contribute and helping developers leverage advanced AI without high costs and expertise requirements. However, the current prompt marketplaces are centralised in nature where the prompts are stored in the company's servers.
Why are people buying prompts instead of the images themselves?
People are buying prompts instead of the images themselves because prompts provide them with the flexibility to modify and curate the generated image according to their specific needs. By purchasing prompts, individuals gain inspiration and a starting point to create unique and customized images that align with their creative vision. Prompt modification allows users to build upon existing ideas and tailor them to their specific requirements, offering a more personalized and curated outcome.
Problems that customers are facing.
Currently, there is no way for consumers to verify if the purchased prompt has the potential to generate the corresponding image. This lack of verification opens the door for prompt listers to potentially deceive users by presenting a random prompt alongside appealing images, leading to potential scams.
Our Approach
This project aims to mitigate the risk of centralization by enabling creators to generate stable mid-journey creations on the client side (we couldn't do it client side within the hackathon deadline and do it on a server but plan to improve it after the hackathon), along with a ZK proof. Once they upload their creation to the marketplace, the proof is verified. If a buyer is interested in a creator's work, they can easily verify the ZK proof, which confirms that the creation was generated using the real prompt. We utilized the EZKL library for verifiable inference to create an attestation proof, converting the .onyx file generated from the Pytorch model to a Rust-based Halo2-backed circuit.