2 ms·

How can text-to-image models be patched against the vulnerability that the attack in this paper exploits?

• smoldesu 4y ago
  Training with older unpoisoned datasets seems like a good start.