Quoted By:
I did some testing with training with very few images (15)
X-axis: embed trained for 30k steps with 16 vectors, same embed at 10k steps, 10k steps with only 4 vectors
Y-axis: No hypernetwork, hypernetwork trained on the same 15 pics for 10k steps
What's the takeaway here? I don't know, you tell me. Tomorrow I'll do one with 100 or so images instead and see how it compares