Paul
8ba8e915ed
This looks like a typo where Union is being imported from `git` instead of `typing` like elsewhere in the repo. I do not have GitPython installed in my container, so this was erroring: ```from git import Union ModuleNotFoundError: No module named 'git'```
256 lines
12 KiB
Python
256 lines
12 KiB
Python
from typing import Union
|
|
from torch import Tensor
|
|
|
|
from .freeinit import FreeInitFilter
|
|
from .sample_settings import (FreeInitOptions, IterationOptions,
|
|
NoiseLayerAdd, NoiseLayerAddWeighted, NoiseLayerGroup, NoiseLayerReplace, NoiseLayerType,
|
|
SeedNoiseGeneration, SampleSettings, CustomCFGKeyframeGroup, CustomCFGKeyframe)
|
|
from .utils_model import BIGMIN, BIGMAX, SigmaSchedule
|
|
|
|
|
|
class SampleSettingsNode:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {
|
|
"required": {
|
|
"batch_offset": ("INT", {"default": 0, "min": 0, "max": BIGMAX}),
|
|
"noise_type": (NoiseLayerType.LIST,),
|
|
"seed_gen": (SeedNoiseGeneration.LIST,),
|
|
"seed_offset": ("INT", {"default": 0, "min": BIGMIN, "max": BIGMAX}),
|
|
},
|
|
"optional": {
|
|
"noise_layers": ("NOISE_LAYERS",),
|
|
"iteration_opts": ("ITERATION_OPTS",),
|
|
"seed_override": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff, "forceInput": True}),
|
|
"adapt_denoise_steps": ("BOOLEAN", {"default": False},),
|
|
"custom_cfg": ("CUSTOM_CFG",),
|
|
"sigma_schedule": ("SIGMA_SCHEDULE",),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("SAMPLE_SETTINGS",)
|
|
RETURN_NAMES = ("settings",)
|
|
CATEGORY = "Animate Diff 🎭🅐🅓"
|
|
FUNCTION = "create_settings"
|
|
|
|
def create_settings(self, batch_offset: int, noise_type: str, seed_gen: str, seed_offset: int, noise_layers: NoiseLayerGroup=None,
|
|
iteration_opts: IterationOptions=None, seed_override: int=None, adapt_denoise_steps=False,
|
|
custom_cfg: CustomCFGKeyframeGroup=None, sigma_schedule: SigmaSchedule=None):
|
|
sampling_settings = SampleSettings(batch_offset=batch_offset, noise_type=noise_type, seed_gen=seed_gen, seed_offset=seed_offset, noise_layers=noise_layers,
|
|
iteration_opts=iteration_opts, seed_override=seed_override, adapt_denoise_steps=adapt_denoise_steps,
|
|
custom_cfg=custom_cfg, sigma_schedule=sigma_schedule)
|
|
return (sampling_settings,)
|
|
|
|
|
|
class NoiseLayerReplaceNode:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {
|
|
"required": {
|
|
"batch_offset": ("INT", {"default": 0, "min": 0, "max": BIGMAX}),
|
|
"noise_type": (NoiseLayerType.LIST,),
|
|
"seed_gen_override": (SeedNoiseGeneration.LIST_WITH_OVERRIDE,),
|
|
"seed_offset": ("INT", {"default": 0, "min": BIGMIN, "max": BIGMAX}),
|
|
},
|
|
"optional": {
|
|
"prev_noise_layers": ("NOISE_LAYERS",),
|
|
"mask_optional": ("MASK",),
|
|
"seed_override": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff, "forceInput": True}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("NOISE_LAYERS",)
|
|
CATEGORY = "Animate Diff 🎭🅐🅓/noise layers"
|
|
FUNCTION = "create_layers"
|
|
|
|
def create_layers(self, batch_offset: int, noise_type: str, seed_gen_override: str, seed_offset: int,
|
|
prev_noise_layers: NoiseLayerGroup=None, mask_optional: Tensor=None, seed_override: int=None,):
|
|
# prepare prev_noise_layers
|
|
if prev_noise_layers is None:
|
|
prev_noise_layers = NoiseLayerGroup()
|
|
prev_noise_layers = prev_noise_layers.clone()
|
|
# create layer
|
|
layer = NoiseLayerReplace(noise_type=noise_type, batch_offset=batch_offset, seed_gen_override=seed_gen_override, seed_offset=seed_offset,
|
|
seed_override=seed_override, mask=mask_optional)
|
|
prev_noise_layers.add_to_start(layer)
|
|
return (prev_noise_layers,)
|
|
|
|
|
|
class NoiseLayerAddNode:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {
|
|
"required": {
|
|
"batch_offset": ("INT", {"default": 0, "min": 0, "max": BIGMAX}),
|
|
"noise_type": (NoiseLayerType.LIST,),
|
|
"seed_gen_override": (SeedNoiseGeneration.LIST_WITH_OVERRIDE,),
|
|
"seed_offset": ("INT", {"default": 0, "min": BIGMIN, "max": BIGMAX}),
|
|
"noise_weight": ("FLOAT", {"default": 0.5, "min": 0.0, "max": 10.0, "step": 0.001}),
|
|
},
|
|
"optional": {
|
|
"prev_noise_layers": ("NOISE_LAYERS",),
|
|
"mask_optional": ("MASK",),
|
|
"seed_override": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff, "forceInput": True}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("NOISE_LAYERS",)
|
|
CATEGORY = "Animate Diff 🎭🅐🅓/noise layers"
|
|
FUNCTION = "create_layers"
|
|
|
|
def create_layers(self, batch_offset: int, noise_type: str, seed_gen_override: str, seed_offset: int,
|
|
noise_weight: float,
|
|
prev_noise_layers: NoiseLayerGroup=None, mask_optional: Tensor=None, seed_override: int=None,):
|
|
# prepare prev_noise_layers
|
|
if prev_noise_layers is None:
|
|
prev_noise_layers = NoiseLayerGroup()
|
|
prev_noise_layers = prev_noise_layers.clone()
|
|
# create layer
|
|
layer = NoiseLayerAdd(noise_type=noise_type, batch_offset=batch_offset, seed_gen_override=seed_gen_override, seed_offset=seed_offset,
|
|
seed_override=seed_override, mask=mask_optional,
|
|
noise_weight=noise_weight)
|
|
prev_noise_layers.add_to_start(layer)
|
|
return (prev_noise_layers,)
|
|
|
|
|
|
class NoiseLayerAddWeightedNode:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {
|
|
"required": {
|
|
"batch_offset": ("INT", {"default": 0, "min": 0, "max": BIGMAX}),
|
|
"noise_type": (NoiseLayerType.LIST,),
|
|
"seed_gen_override": (SeedNoiseGeneration.LIST_WITH_OVERRIDE,),
|
|
"seed_offset": ("INT", {"default": 0, "min": BIGMIN, "max": BIGMAX}),
|
|
"noise_weight": ("FLOAT", {"default": 0.5, "min": 0.0, "max": 10.0, "step": 0.001}),
|
|
"balance_multiplier": ("FLOAT", {"default": 1.0, "min": 0.0, "step": 0.001}),
|
|
},
|
|
"optional": {
|
|
"prev_noise_layers": ("NOISE_LAYERS",),
|
|
"mask_optional": ("MASK",),
|
|
"seed_override": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff, "forceInput": True}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("NOISE_LAYERS",)
|
|
CATEGORY = "Animate Diff 🎭🅐🅓/noise layers"
|
|
FUNCTION = "create_layers"
|
|
|
|
def create_layers(self, batch_offset: int, noise_type: str, seed_gen_override: str, seed_offset: int,
|
|
noise_weight: float, balance_multiplier: float,
|
|
prev_noise_layers: NoiseLayerGroup=None, mask_optional: Tensor=None, seed_override: int=None,):
|
|
# prepare prev_noise_layers
|
|
if prev_noise_layers is None:
|
|
prev_noise_layers = NoiseLayerGroup()
|
|
prev_noise_layers = prev_noise_layers.clone()
|
|
# create layer
|
|
layer = NoiseLayerAddWeighted(noise_type=noise_type, batch_offset=batch_offset, seed_gen_override=seed_gen_override, seed_offset=seed_offset,
|
|
seed_override=seed_override, mask=mask_optional,
|
|
noise_weight=noise_weight, balance_multiplier=balance_multiplier)
|
|
prev_noise_layers.add_to_start(layer)
|
|
return (prev_noise_layers,)
|
|
|
|
|
|
class IterationOptionsNode:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {
|
|
"required": {
|
|
"iterations": ("INT", {"default": 1, "min": 1}),
|
|
},
|
|
"optional": {
|
|
"iter_batch_offset": ("INT", {"default": 0, "min": 0, "max": BIGMAX}),
|
|
"iter_seed_offset": ("INT", {"default": 0, "min": BIGMIN, "max": BIGMAX}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("ITERATION_OPTS",)
|
|
CATEGORY = "Animate Diff 🎭🅐🅓/iteration opts"
|
|
FUNCTION = "create_iter_opts"
|
|
|
|
def create_iter_opts(self, iterations: int, iter_batch_offset: int=0, iter_seed_offset: int=0):
|
|
iter_opts = IterationOptions(iterations=iterations, iter_batch_offset=iter_batch_offset, iter_seed_offset=iter_seed_offset)
|
|
return (iter_opts,)
|
|
|
|
|
|
class FreeInitOptionsNode:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {
|
|
"required": {
|
|
"iterations": ("INT", {"default": 2, "min": 1}),
|
|
"filter": (FreeInitFilter.LIST,),
|
|
"d_s": ("FLOAT", {"default": 0.25, "min": 0.0, "max": 1.0, "step": 0.001}),
|
|
"d_t": ("FLOAT", {"default": 0.25, "min": 0.0, "max": 1.0, "step": 0.001}),
|
|
"n_butterworth": ("INT", {"default": 4, "min": 1, "max": 100},),
|
|
"sigma_step": ("INT", {"default": 999, "min": 1, "max": 999}),
|
|
"apply_to_1st_iter": ("BOOLEAN", {"default": False}),
|
|
"init_type": (FreeInitOptions.LIST,)
|
|
},
|
|
"optional": {
|
|
"iter_batch_offset": ("INT", {"default": 0, "min": 0, "max": BIGMAX}),
|
|
"iter_seed_offset": ("INT", {"default": 1, "min": BIGMIN, "max": BIGMAX}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("ITERATION_OPTS",)
|
|
CATEGORY = "Animate Diff 🎭🅐🅓/iteration opts"
|
|
FUNCTION = "create_iter_opts"
|
|
|
|
def create_iter_opts(self, iterations: int, filter: str, d_s: float, d_t: float, n_butterworth: int,
|
|
sigma_step: int, apply_to_1st_iter: bool, init_type: str,
|
|
iter_batch_offset: int=0, iter_seed_offset: int=1):
|
|
# init_type does nothing for now, not until I add more methods of applying low+high freq noise
|
|
iter_opts = FreeInitOptions(iterations=iterations, step=sigma_step, apply_to_1st_iter=apply_to_1st_iter,
|
|
filter=filter, d_s=d_s, d_t=d_t, n=n_butterworth, init_type=init_type,
|
|
iter_batch_offset=iter_batch_offset, iter_seed_offset=iter_seed_offset)
|
|
return (iter_opts,)
|
|
|
|
|
|
class CustomCFGNode:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {
|
|
"required": {
|
|
"cfg_multival": ("MULTIVAL",),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("CUSTOM_CFG",)
|
|
CATEGORY = "Animate Diff 🎭🅐🅓/sample settings"
|
|
FUNCTION = "create_custom_cfg"
|
|
|
|
def create_custom_cfg(self, cfg_multival: Union[float, Tensor]):
|
|
keyframe = CustomCFGKeyframe(cfg_multival=cfg_multival)
|
|
cfg_custom = CustomCFGKeyframeGroup()
|
|
cfg_custom.add(keyframe)
|
|
return (cfg_custom,)
|
|
|
|
|
|
class CustomCFGKeyframeNode:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return {
|
|
"required": {
|
|
"cfg_multival": ("MULTIVAL",),
|
|
"start_percent": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 1.0, "step": 0.001}),
|
|
"guarantee_steps": ("INT", {"default": 1, "min": 0, "max": BIGMAX}),
|
|
},
|
|
"optional": {
|
|
"prev_custom_cfg": ("CUSTOM_CFG",),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("CUSTOM_CFG",)
|
|
CATEGORY = "Animate Diff 🎭🅐🅓/sample settings"
|
|
FUNCTION = "create_custom_cfg"
|
|
|
|
def create_custom_cfg(self, cfg_multival: Union[float, Tensor], start_percent: float=0.0, guarantee_steps: int=1,
|
|
prev_custom_cfg: CustomCFGKeyframeGroup=None):
|
|
if not prev_custom_cfg:
|
|
prev_custom_cfg = CustomCFGKeyframeGroup()
|
|
prev_custom_cfg = prev_custom_cfg.clone()
|
|
keyframe = CustomCFGKeyframe(cfg_multival=cfg_multival, start_percent=start_percent, guarantee_steps=guarantee_steps)
|
|
prev_custom_cfg.add(keyframe)
|
|
return (prev_custom_cfg,)
|