johnowhitaker/pyramid_noise_test_600steps_08discount

A model trained with Pyramid Noise - see https://wandb.ai/johnowhitaker/multires_noise/reports/Multi-Resolution-Noise-for-Diffusion-Model-Training--VmlldzozNjYyOTU2 for details

from torch import nn
import random

def pyramid_noise_like(x, discount=0.8):
  b, c, w, h = x.shape
  u = nn.Upsample(size=(w, h), mode='bilinear')
  noise = torch.randn_like(x)
  for i in range(6):
    r = random.random()*2+2 # Rather than always going 2x, 
    w, h = max(1, int(w/(r**i))), max(1, int(h/(r**i)))
    noise += u(torch.randn(b, c, w, h).to(x)) * discount**i
    if w==1 or h==1: break 
  return noise / noise.std() # Scale back to unit variance

To use the mode for inference, just load it like a normal stable diffusion pipeline:

from diffusers import StableDiffusionPipeline

model_path = "johnowhitaker/pyramid_noise_test_600steps_08discount"
pipe = StableDiffusionPipeline.from_pretrained(model_path, torch_dtype=torch.float16)
pipe.to("cuda")

image = pipe(prompt="A black image").images[0]
image