About the Project

Inspiration

My younger sibling is an active traditional Korean painter living in a rural area. Like many emerging artists, they face intense isolation and financial pressure.
I wanted to use modern technology to help my sibling overcome these challenges.
People often say that an artist’s work gains value after they’re gone — but I believe it should have value while they're still alive.

What It Does

1. Copyright Protection Services

Invisible Watermark
Embeds copyright information into the image that is invisible to the human eye. If the image is leaked illegally, this technology allows us to trace the source.

Adversarial Noise
Applies imperceptible noise to select pixels in the image to prevent AI models from easily using it for training or imitation.

2. Web Crawling & Image Similarity Detection

On-demand image crawling and similarity analysis across the web to detect possible infringements using:

CLIP
Measures style and composition similarity using image-image and text-image embeddings. Helps answer questions like: "Does this AI-generated image mimic the style of this artist?"
DISTS
Evaluates both structural layout and texture, making it well-suited for detecting stylistic plagiarism: "Has the artist’s unique style been replicated?"
LPIPS
A perceptual similarity metric aligned with human vision. Effective as a first filter in plagiarism detection due to its human-like assessment of visual similarity.
Grad-CAM
Highlights the areas a model focuses on when classifying an image, helping identify whether key visual features were replicated.

How We Built It

We started with open-source AI models and image similarity algorithms, adapting them to the unique challenges of copyright protection. We built a multi-agent system where different models (CLIP, DISTS, LPIPS, Grad-CAM) cooperate to assess visual similarity and plagiarism risk.
The system also includes a web scraping tool that collects publicly available artworks from the internet for comparison. We integrated adversarial learning techniques and watermark embedding to strengthen protection mechanisms. The platform was built using Python, PyTorch, FastAPI, and cloud services like GCP and Firebase.

Challenges We Ran Into

Integrating AI Technologies
While effective, adapting them to large language models like ChatGPT and making them run reliably proved difficult. Creating interpretable results that could be used in legal or dispute scenarios required multiple iterations.
Limited Hardware
Development was constrained by hardware limitations — frequent crashes due to memory shortages on personal laptops. Training and inference of vision models like CLIP required optimization to run efficiently on consumer-grade GPUs.
Legal and Ethical Complexity
Understanding how to frame AI-generated similarity scores as usable evidence for copyright claims required consultation with legal and artistic communities.

Accomplishments That We're Proud Of

Successfully implemented a multi-layer AI system capable of analyzing artistic similarity with human-like perception.
Developed a working prototype of the copyright detection workflow used by real artists.
Gained interest and feedback from professional painters and illustrators in Korea.
Built an end-to-end demo including NFT minting and royalty tracking using smart contracts.

What We Learned

Artists need more than just "blockchain": they need protection, recognition, and control over their works in the AI era.
It’s not just about building with AI — it’s about making AI trustworthy and explainable for human use.
Collaboration between art and tech requires empathy, not just engineering.

What's Next for ArtNest

Launch beta with real artist users and gather feedback on copyright risk detection
Enable automated PDF reports for suspected infringements (usable for legal action)
Expand to support video and 3D artwork protection
Integrate with major NFT marketplaces to enforce resale royalties
Form partnerships with creator communities, art schools, and cultural institutions