Project Overview: The Object-Key System

The Elevator Pitch Demo

Imagine you need to log into your Bank account. Instead of typing "Bank123!", you open our app, take a photo of your favorite coffee mug, and specify "Bank." Our system instantly generates a 10-character, high-entropy password.

When you return to the site tomorrow, you take a photo of that same mug again. Our AI recognizes the semantic essence of the object—regardless of the lighting or angle—and regenerates that exact same password. If you try to use a picture of a stapler or a shoe, the system rejects it.

You remember the object; the AI handles the security.

How the Magic Works (The Technical Demo)

When you upload that photo, three distinct AI processes happen in milliseconds:

  1. Isolation (YOLO): Our backend receives your image and uses the YOLO (You Only Look Once) model to find the most prominent object. It "crops out" the background noise so the AI is not confused by your messy desk.
  2. Vectorization (DinoV2): The cropped object is fed into DinoV2. This model does not just look for colors or shapes; it understands the object’s structure. It turns the image into a unique mathematical "fingerprint"—a high-dimensional vector v.
  3. Cryptographic Locking: We take that vector v, add a secret "salt" s (a random string for extra security), and the name of the app L. We hash them together: P = H(v + s, L). The result P is the password that gets typed into your login field.

Why this is a Breakthrough for Cybersecurity

  • AI vs. AI: Traditional password managers are susceptible to social engineering. Our system uses AI to verify your physical environment, creating a "contextual lock" that a remote hacker cannot replicate.
  • The "Zero-Memory" Burden: You do not need to learn a password manager interface. You just need to look at an object you interact with daily.
  • Database Safety: Even if a company’s database is hacked, the attacker only sees scrambled hashes. Without the physical object to "re-generate" the vector v, the passwords remain useless.

Ready to see it in action?

You can experiment with the current version hosted on our Hugging Face Space.

  • To demo: Just upload a clear photo of an object (like a water bottle) and see how the system identifies and fingerprints it.
  • A note on security: As this is a hackathon prototype, please use it for testing purposes—we are still finalizing our database security protocols before moving to production-grade encryption.

Built With

Share this project:

Updates