45 lines
1.5 KiB
Python
45 lines
1.5 KiB
Python
import torch
|
|
|
|
|
|
class ModelSamplerTonemapNoiseTest:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {"required": { "model": ("MODEL",),
|
|
"multiplier": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step": 0.01}),
|
|
}}
|
|
RETURN_TYPES = ("MODEL",)
|
|
FUNCTION = "patch"
|
|
|
|
CATEGORY = "custom_node_experiments"
|
|
|
|
def patch(self, model, multiplier):
|
|
|
|
def sampler_tonemap_reinhard(args):
|
|
cond = args["cond"]
|
|
uncond = args["uncond"]
|
|
cond_scale = args["cond_scale"]
|
|
noise_pred = (cond - uncond)
|
|
noise_pred_vector_magnitude = (torch.linalg.vector_norm(noise_pred, dim=(1)) + 0.0000000001)[:,None]
|
|
noise_pred /= noise_pred_vector_magnitude
|
|
|
|
mean = torch.mean(noise_pred_vector_magnitude, dim=(1,2,3), keepdim=True)
|
|
std = torch.std(noise_pred_vector_magnitude, dim=(1,2,3), keepdim=True)
|
|
|
|
top = (std * 3 + mean) * multiplier
|
|
|
|
#reinhard
|
|
noise_pred_vector_magnitude *= (1.0 / top)
|
|
new_magnitude = noise_pred_vector_magnitude / (noise_pred_vector_magnitude + 1.0)
|
|
new_magnitude *= top
|
|
|
|
return uncond + noise_pred * new_magnitude * cond_scale
|
|
|
|
m = model.clone()
|
|
m.set_model_sampler_cfg_function(sampler_tonemap_reinhard)
|
|
return (m, )
|
|
|
|
|
|
NODE_CLASS_MAPPINGS = {
|
|
"ModelSamplerTonemapNoiseTest": ModelSamplerTonemapNoiseTest,
|
|
}
|