I have read that typically, training stable diffusion models uses EDM.
So when I saw this, I thought, "sounds like a great new efficient way to train models."
However, the examples Ive seen so far for this, use a new custom model.
Can this code be used to fine-tune pre-existing models, that previouisly used EDM, without changing the architecture of the model?
Or does EDM2 only work on these brand new architectural models?
I have read that typically, training stable diffusion models uses EDM.
So when I saw this, I thought, "sounds like a great new efficient way to train models."
However, the examples Ive seen so far for this, use a new custom model.
Can this code be used to fine-tune pre-existing models, that previouisly used EDM, without changing the architecture of the model?
Or does EDM2 only work on these brand new architectural models?