File size: 10,960 Bytes
3e5f61c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 | from abc import ABC, abstractmethod
from typing import Union
import torch
import torch.nn.functional as F
from torch import Tensor, nn
import comfy.model_management as model_management
import comfy.ops
from comfy.cli_args import args
from comfy.ldm.modules.attention import attention_basic, attention_pytorch, attention_split, attention_sub_quad, default
from comfy.controlnet import broadcast_image_to
from comfy.utils import repeat_to_batch_size
from .motion_lora import MotionLoraInfo
from .logger import logger
# until xformers bug is fixed, do not use xformers for VersatileAttention! TODO: change this when fix is out
# logic for choosing optimized_attention method taken from comfy/ldm/modules/attention.py
optimized_attention_mm = attention_basic
if model_management.xformers_enabled():
pass
#optimized_attention_mm = attention_xformers
if model_management.pytorch_attention_enabled():
optimized_attention_mm = attention_pytorch
else:
if args.use_split_cross_attention:
optimized_attention_mm = attention_split
else:
optimized_attention_mm = attention_sub_quad
# maintain backwards compatibility with the comfy.ops hasattr check (TODO: remove once a non-backwards compatible change happens)
class CrossAttentionMM(nn.Module):
def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0., dtype=None, device=None,
operations=comfy.ops.disable_weight_init if hasattr(comfy.ops, "disable_weight_init") else comfy.ops):
super().__init__()
inner_dim = dim_head * heads
context_dim = default(context_dim, query_dim)
self.heads = heads
self.dim_head = dim_head
self.scale = None
self.default_scale = dim_head ** -0.5
self.to_q = operations.Linear(query_dim, inner_dim, bias=False, dtype=dtype, device=device)
self.to_k = operations.Linear(context_dim, inner_dim, bias=False, dtype=dtype, device=device)
self.to_v = operations.Linear(context_dim, inner_dim, bias=False, dtype=dtype, device=device)
self.to_out = nn.Sequential(operations.Linear(inner_dim, query_dim, dtype=dtype, device=device), nn.Dropout(dropout))
def forward(self, x, context=None, value=None, mask=None, scale_mask=None):
q = self.to_q(x)
context = default(context, x)
k: Tensor = self.to_k(context)
if value is not None:
v = self.to_v(value)
del value
else:
v = self.to_v(context)
# apply custom scale by multiplying k by scale factor
if self.scale is not None:
k *= self.scale
# apply scale mask, if present
if scale_mask is not None:
k *= scale_mask
out = optimized_attention_mm(q, k, v, self.heads, mask)
return self.to_out(out)
# super class to TemporalTransformer-like classes
class TemporalTransformerGeneric:
def temporal_transformer_init(self, default_length: int):
self.video_length = default_length
self.full_length = default_length
self.scale_min = 1.0
self.scale_max = 1.0
self.raw_scale_mask: Union[Tensor, None] = None
self.temp_scale_mask: Union[Tensor, None] = None
self.sub_idxs: Union[list[int], None] = None
self.prev_hidden_states_batch = 0
def reset_temp_vars(self):
del self.temp_scale_mask
self.temp_scale_mask = None
self.prev_hidden_states_batch = 0
def get_scale_mask(self, hidden_states: Tensor) -> Union[Tensor, None]:
# if no raw mask, return None
if self.raw_scale_mask is None:
return None
shape = hidden_states.shape
batch, channel, height, width = shape
# if temp mask already calculated, return it
if self.temp_scale_mask != None:
# check if hidden_states batch matches
if batch == self.prev_hidden_states_batch:
if self.sub_idxs is not None:
return self.temp_scale_mask[:, self.sub_idxs, :]
return self.temp_scale_mask
# if does not match, reset cached temp_scale_mask and recalculate it
del self.temp_scale_mask
self.temp_scale_mask = None
# otherwise, calculate temp mask
self.prev_hidden_states_batch = batch
mask = prepare_mask_batch(self.raw_scale_mask, shape=(self.full_length, 1, height, width))
mask = repeat_to_batch_size(mask, self.full_length)
# if mask not the same amount length as full length, make it match
if self.full_length != mask.shape[0]:
mask = broadcast_image_to(mask, self.full_length, 1)
# reshape mask to attention K shape (h*w, latent_count, 1)
batch, channel, height, width = mask.shape
# first, perform same operations as on hidden_states,
# turning (b, c, h, w) -> (b, h*w, c)
mask = mask.permute(0, 2, 3, 1).reshape(batch, height*width, channel)
# then, make it the same shape as attention's k, (h*w, b, c)
mask = mask.permute(1, 0, 2)
# make masks match the expected length of h*w
batched_number = shape[0] // self.video_length
if batched_number > 1:
mask = torch.cat([mask] * batched_number, dim=0)
# cache mask and set to proper device
self.temp_scale_mask = mask
# move temp_scale_mask to proper dtype + device
self.temp_scale_mask = self.temp_scale_mask.to(dtype=hidden_states.dtype, device=hidden_states.device)
# return subset of masks, if needed
if self.sub_idxs is not None:
return self.temp_scale_mask[:, self.sub_idxs, :]
return self.temp_scale_mask
class BlockType:
UP = "up"
DOWN = "down"
MID = "mid"
class GenericMotionWrapper(nn.Module, ABC):
def __init__(self, mm_hash: str, mm_name: str, loras: list[MotionLoraInfo]):
super().__init__()
self.down_blocks: nn.ModuleList = None
self.up_blocks: nn.ModuleList = None
self.mid_block = None
self.mm_hash = mm_hash
self.mm_name = mm_name
self.version = "FILLTHISIN"
self.injector_version = "VERYIMPORTANT_FILLTHISIN"
self.AD_video_length: int = 0
self.loras = loras
def has_loras(self) -> bool:
# TODO: fix this to return False if has an empty list as well
# but only after implementing a fix for lowvram loading
return self.loras is not None
@abstractmethod
def set_video_length(self, video_length: int, full_length: int):
pass
@abstractmethod
def set_scale_multiplier(self, multiplier: Union[float, None]):
pass
@abstractmethod
def set_masks(self, masks: Tensor, min_val: float, max_val: float):
pass
@abstractmethod
def set_sub_idxs(self, sub_idxs: list[int]):
pass
@abstractmethod
def reset_temp_vars(self):
pass
def reset_scale_multiplier(self):
self.set_scale_multiplier(None)
def reset_sub_idxs(self):
self.set_sub_idxs(None)
def reset(self):
self.reset_sub_idxs()
self.reset_scale_multiplier()
self.reset_temp_vars()
class GroupNormAD(torch.nn.GroupNorm):
def __init__(self, num_groups: int, num_channels: int, eps: float = 1e-5, affine: bool = True,
device=None, dtype=None) -> None:
super().__init__(num_groups=num_groups, num_channels=num_channels, eps=eps, affine=affine, device=device, dtype=dtype)
def forward(self, input: Tensor) -> Tensor:
return F.group_norm(
input, self.num_groups, self.weight, self.bias, self.eps)
# applies min-max normalization, from:
# https://stackoverflow.com/questions/68791508/min-max-normalization-of-a-tensor-in-pytorch
def normalize_min_max(x: Tensor, new_min = 0.0, new_max = 1.0):
x_min, x_max = x.min(), x.max()
return (((x - x_min)/(x_max - x_min)) * (new_max - new_min)) + new_min
# adapted from comfy/sample.py
def prepare_mask_batch(mask: Tensor, shape: Tensor, multiplier: int=1, match_dim1=False):
mask = mask.clone()
mask = torch.nn.functional.interpolate(mask.reshape((-1, 1, mask.shape[-2], mask.shape[-1])), size=(shape[2]*multiplier, shape[3]*multiplier), mode="bilinear")
if match_dim1:
mask = torch.cat([mask] * shape[1], dim=1)
return mask
class NoiseType:
DEFAULT = "default"
REPEATED = "repeated"
CONSTANT = "constant"
AUTO1111 = "auto1111"
LIST = [DEFAULT, REPEATED, CONSTANT, AUTO1111]
@classmethod
def prepare_noise(cls, noise_type: str, latents: Tensor, noise: Tensor, context_length: int, seed: int):
if noise_type == cls.DEFAULT:
return noise
elif noise_type == cls.REPEATED:
return cls.prepare_noise_repeated(latents, noise, context_length, seed)
elif noise_type == cls.CONSTANT:
return cls.prepare_noise_constant(latents, noise, context_length, seed)
elif noise_type == cls.AUTO1111:
return cls.prepare_noise_auto1111(latents, noise, context_length, seed)
logger.warning(f"Noise type {noise_type} not recognized, proceeding with default noise.")
return noise
@classmethod
def prepare_noise_repeated(cls, latents: Tensor, noise: Tensor, context_length: int, seed: int):
if not context_length:
return noise
length = latents.shape[0]
generator = torch.manual_seed(seed)
noise = torch.randn(latents.size(), dtype=latents.dtype, layout=latents.layout, generator=generator, device="cpu")
noise_set = noise[:context_length]
cat_count = (length // context_length) + 1
noise_set = torch.cat([noise_set] * cat_count, dim=0)
noise_set = noise_set[:length]
return noise_set
@classmethod
def prepare_noise_constant(cls, latents: Tensor, noise: Tensor, context_length: int, seed: int):
length = latents.shape[0]
single_shape = (1, latents.shape[1], latents.shape[2], latents.shape[3])
generator = torch.manual_seed(seed)
noise = torch.randn(single_shape, dtype=latents.dtype, layout=latents.layout, generator=generator, device="cpu")
return torch.cat([noise] * length, dim=0)
@classmethod
def prepare_noise_auto1111(cls, latents: Tensor, noise: Tensor, context_length: int, seed: int):
# auto1111 applies growing seeds for a batch
length = latents.shape[0]
single_shape = (1, latents.shape[1], latents.shape[2], latents.shape[3])
all_noises = []
# i starts at 0
for i in range(length):
generator = torch.manual_seed(seed+i)
all_noises.append(torch.randn(single_shape, dtype=latents.dtype, layout=latents.layout, generator=generator, device="cpu"))
return torch.cat(all_noises, dim=0)
class MotionCompatibilityError(ValueError):
pass
|