Built on a foundation of transparency.

We believe in building in the open. Our modular architecture is designed to scale without bloat, allowing us to roll out high-performance features at no additional cost to you. This isn't just a tool; it's an ecosystem built by creators, for creators. If a feature can save you time or grow your revenue, we'll build it into our core.

1 shipped2 in progress0 requests

AI poisoning and visual fingerprinting

Last updated Jun 3

Protect your images before they're ever seen.

The moment you upload, two things happen invisibly:

1. Adversarial Poisoning (Glaze/Nightshade)

Your image is treated with a layer of adversarial protection—microscopic changes imperceptible to the human eye that cause AI training models to misread your work. If someone tries to train an AI on your images, it learns the wrong thing.

  • How: The system integrates the Glaze or Nightshade CLI/API (developed by the University of Chicago). It applies microscopic pixel shifts that trick AI training models into misinterpreting the image's style or content.

2. Perceptual DNA Fingerprinting (DINOv2)

Your image receives a visual fingerprint—a unique signature based on what the image actually looks like, not just its file data. Unlike a standard checksum, this fingerprint survives cropping, screenshotting, color adjustments, and watermark removal. It is how we recognize your image even when someone has tried to alter it.

  • How: The system utilizes Meta’s DINOv2 model to generate a 384-dimensional vector representation of the asset.
  • Why DINOv2? Unlike standard perceptual hashing (pHash), DINOv2 evaluates high-level visual features. If a bad actor takes a screenshot of the image, crops out the watermark, and flips it horizontally, the resulting vector remains 98% identical.
  • Storage Infrastructure: Vector embeddings are indexed inside a dedicated vector database (such as Pinecone or Qdrant) to power sub-second, real-time "Reverse DNA" copyright lookups.