No, it took a week to refine the attack algorithm, the collision generation itself is fast
The point of perceptual hashes is to let you check if two things are similar enough after transformations like scaling and reencoding, so you can't rely on that here