Inspiration

In an era where AI-generated imagery is increasingly indistinguishable from photographs taken in the real world, the question of authenticity has never been more pertinent. As we can no longer rely solely on traditional metadata, the quest for innovative methods to certify the provenance of visual content is paramount.

What it does

"The Mask" is a pioneering technology that discerns AI-generated images from genuine photographs by incorporating a dual-layer verification system. It operates not only at the model level during the image generation process but also applies a post-processing verification layer to each image. This system utilizes a unique cryptographic hash from the Solana blockchain, embedding it as an immutable watermark within the image's data.

How we built it(differently for this hackathon)

For this hackathon, we shifted our focus from embedding watermarks at the model level to a more universal application that can be applied to images generated by any model. Here's how we adapted our approach:

Universal Watermarking Algorithm: We developed a sophisticated algorithm capable of applying a unique watermark to images regardless of the generative model used. This algorithm analyzes the image's content to find the most unobtrusive area to embed the watermark.

Post-Processing Integration: After an image is generated by any AI model, our system takes over in the post-processing phase. It seamlessly integrates the blockchain-derived hash into the image's data without affecting the visual quality.

Model Agnostic Protocol: Our protocol is designed to be independent of the generative model's architecture. Whether the image is produced by latent diffusion or any other AI technique, our watermarking process remains consistent and effective.

Blockchain Hash Retrieval and Embedding: We created a service that quickly retrieves a hash from the Solana blockchain and embeds it into the image. This service is optimized for high throughput and low latency, ensuring it can operate at scale.

Verification Layer: Alongside the watermarking, we implemented a verification layer that can detect and decode the watermark from any image. This ensures that the authenticity of an image can be verified long after its creation, regardless of its journey across the internet.

By focusing on a post-model watermarking system, we've ensured that "The Mask" is not only a solution for images generated by a specific AI model but a universal standard for digital image verification.

Challenges we ran into

Uploading using the browser is not something to be proud of.

Accomplishments that we're proud of

We kinda did it!

What's next for The Mask

(stay tuned)

Built With

Share this project:

Updates