stable-diffusion-webui-forge/modules/sd_vae_taesd.py
2026-05-26 14:25:28 +08:00

277 lines
11 KiB
Python

# Tiny AutoEncoder for Stable Diffusion
# https://github.com/madebyollin/taesd/blob/main/taesd.py
# Tiny AutoEncoder for Hunyuan Video
# https://github.com/madebyollin/taehv/blob/main/taehv.py
# reference:
# - https://github.com/Comfy-Org/ComfyUI/blob/v0.21.0/comfy/taesd/taehv.py
import os
from typing import TYPE_CHECKING
if TYPE_CHECKING:
from modules_forge.packages.huggingface_guess.latent import LatentFormat
from collections import deque, namedtuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from backend.state_dict import load_state_dict
from backend.utils import load_torch_file
from modules import devices, paths_internal, shared
URL: str = "https://github.com/madebyollin/taesd/raw/main/"
URL_V: str = "https://github.com/madebyollin/taehv/raw/main/"
sd_vae_taesd_models: dict[str, nn.Module] = {}
TWorkItem = namedtuple("TWorkItem", ("input_tensor", "block_index"))
def conv(n_in, n_out, **kwargs):
return nn.Conv2d(n_in, n_out, 3, padding=1, **kwargs)
class Clamp(nn.Module):
@staticmethod
def forward(x):
return torch.tanh(x / 3) * 3
class Block(nn.Module):
def __init__(self, n_in, n_out, use_midblock_gn=False):
super().__init__()
self.conv = nn.Sequential(conv(n_in, n_out), nn.ReLU(), conv(n_out, n_out), nn.ReLU(), conv(n_out, n_out))
self.skip = nn.Conv2d(n_in, n_out, 1, bias=False) if n_in != n_out else nn.Identity()
self.fuse = nn.ReLU()
self.pool = None
if use_midblock_gn:
conv1x1 = lambda n_in, n_out: nn.Conv2d(n_in, n_out, 1, bias=False)
n_gn = n_in * 4
self.pool = nn.Sequential(conv1x1(n_in, n_gn), nn.GroupNorm(4, n_gn), nn.ReLU(inplace=True), conv1x1(n_gn, n_in))
def forward(self, x):
if self.pool is not None:
x = x + self.pool(x)
return self.fuse(self.conv(x) + self.skip(x))
def decoder(latent_channels=4, use_midblock_gn=False):
mb_kw = dict(use_midblock_gn=use_midblock_gn)
return nn.Sequential(
*(Clamp(), conv(latent_channels, 64), nn.ReLU()),
*(Block(64, 64, **mb_kw), Block(64, 64, **mb_kw), Block(64, 64, **mb_kw), nn.Upsample(scale_factor=2), conv(64, 64, bias=False)),
*(Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False)),
*(Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False)),
*(Block(64, 64), conv(64, 3)),
)
def encoder(latent_channels=4, use_midblock_gn=False):
mb_kw = dict(use_midblock_gn=use_midblock_gn)
return nn.Sequential(
*(conv(3, 64), Block(64, 64)),
*(conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64)),
*(conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64)),
*(conv(64, 64, stride=2, bias=False), Block(64, 64, **mb_kw), Block(64, 64, **mb_kw), Block(64, 64, **mb_kw)),
conv(64, latent_channels),
)
class TAESDDecoder(nn.Module):
def __init__(self, decoder_path: os.PathLike, latent_channels: int):
super().__init__()
self.latent_channels = 32 if latent_channels == 128 else latent_channels
self.decoder = decoder(self.latent_channels, use_midblock_gn=(self.latent_channels == 32))
load_state_dict(self.decoder, load_torch_file(decoder_path))
def forward(self, x: torch.Tensor) -> torch.Tensor:
if self.latent_channels == 32:
x = x.reshape(x.shape[0], self.latent_channels, 2, 2, x.shape[-2], x.shape[-1]).permute(0, 1, 4, 2, 5, 3).reshape(x.shape[0], self.latent_channels, x.shape[-2] * 2, x.shape[-1] * 2)
return self.decoder(x)
class TAESDEncoder(nn.Module):
def __init__(self, encoder_path: os.PathLike, latent_channels: int):
super().__init__()
self.latent_channels = 32 if latent_channels == 128 else latent_channels
self.encoder = encoder(self.latent_channels, use_midblock_gn=(self.latent_channels == 32))
load_state_dict(self.encoder, load_torch_file(encoder_path))
def forward(self, x_sample: torch.Tensor) -> torch.Tensor:
if self.latent_channels == 32:
x_sample = x_sample.reshape(x_sample.shape[0], self.latent_channels, x_sample.shape[-2] // 2, 2, x_sample.shape[-1] // 2, 2).permute(0, 1, 3, 5, 2, 4).reshape(x_sample.shape[0], self.latent_channels * 4, x_sample.shape[-2] // 2, x_sample.shape[-1] // 2)
return self.encoder(x_sample)
class MemBlock(nn.Module):
def __init__(self, n_in, n_out, act_func):
super().__init__()
self.conv = nn.Sequential(conv(n_in * 2, n_out), act_func, conv(n_out, n_out), act_func, conv(n_out, n_out))
self.skip = nn.Conv2d(n_in, n_out, 1, bias=False) if n_in != n_out else nn.Identity()
self.act = act_func
def forward(self, x, past):
return self.act(self.conv(torch.cat([x, past], 1)) + self.skip(x))
class TGrow(nn.Module):
def __init__(self, n_f, stride):
super().__init__()
self.stride = stride
self.conv = nn.Conv2d(n_f, n_f * stride, 1, bias=False)
def forward(self, x):
_, C, H, W = x.shape
x = self.conv(x)
return x.reshape(-1, C, H, W)
class TAEHV(nn.Module):
def __init__(self, latent_channels, decoder_time_upscale=(False, True, True), decoder_space_upscale=(True, True, True)):
super().__init__()
self.image_channels = 3
self.latent_channels = latent_channels
if self.latent_channels == 16: # Wan 2.1
self.patch_size = 1
act_func = nn.ReLU(inplace=True)
else: # HunyuanVideo 1.5
self.patch_size = 2
act_func = nn.LeakyReLU(0.2, inplace=True)
n_f = [256, 128, 64, 64]
self.decoder = nn.Sequential(
*(Clamp(), conv(self.latent_channels, n_f[0]), act_func),
*(MemBlock(n_f[0], n_f[0], act_func), MemBlock(n_f[0], n_f[0], act_func), MemBlock(n_f[0], n_f[0], act_func), nn.Upsample(scale_factor=2 if decoder_space_upscale[0] else 1), TGrow(n_f[0], 2 if decoder_time_upscale[0] else 1), conv(n_f[0], n_f[1], bias=False)),
*(MemBlock(n_f[1], n_f[1], act_func), MemBlock(n_f[1], n_f[1], act_func), MemBlock(n_f[1], n_f[1], act_func), nn.Upsample(scale_factor=2 if decoder_space_upscale[1] else 1), TGrow(n_f[1], 2 if decoder_time_upscale[1] else 1), conv(n_f[1], n_f[2], bias=False)),
*(MemBlock(n_f[2], n_f[2], act_func), MemBlock(n_f[2], n_f[2], act_func), MemBlock(n_f[2], n_f[2], act_func), nn.Upsample(scale_factor=2 if decoder_space_upscale[2] else 1), TGrow(n_f[2], 2 if decoder_time_upscale[2] else 1), conv(n_f[2], n_f[3], bias=False)),
*(act_func, conv(n_f[3], self.image_channels * self.patch_size**2)),
)
self.t_upscale = 2 ** sum(t.stride == 2 for t in self.decoder if isinstance(t, TGrow))
self.frames_to_trim = self.t_upscale - 1
def apply_model_with_memblocks(self, x: torch.Tensor, patch_size: int):
B, T, C, H, W = x.shape
mem = [None] * len(self.decoder)
work_queue = deque([TWorkItem(xt.squeeze(1), 0) for xt in x.chunk(T, dim=1)])
out = []
while work_queue:
xt, i = work_queue.popleft()
if i == len(self.decoder):
if patch_size > 1:
xt = F.pixel_shuffle(xt, patch_size)
out.append(xt)
del xt
else:
b = self.decoder[i]
if isinstance(b, MemBlock):
if mem[i] is None:
xt_new = b(xt, xt * 0)
mem[i] = xt.detach().clone()
else:
xt_new = b(xt, mem[i])
mem[i] = xt.detach().clone()
del xt
work_queue.appendleft(TWorkItem(xt_new, i + 1))
elif isinstance(b, TGrow):
xt = b(xt)
_, C, H, W = xt.shape
for xt_next in reversed(xt.view(B, b.stride * C, H, W).chunk(b.stride, 1)):
work_queue.appendleft(TWorkItem(xt_next, i + 1))
del xt
else:
xt = b(xt)
work_queue.appendleft(TWorkItem(xt, i + 1))
return torch.stack(out, 1)
def decode(self, x: torch.Tensor) -> torch.Tensor:
x = x.movedim(2, 1)
x = self.apply_model_with_memblocks(x, self.patch_size)
return x[:, self.frames_to_trim :]
class TAEHVDecoder(nn.Module):
def __init__(self, decoder_path: os.PathLike, latent_channels: int):
super().__init__()
self.latent_channels = latent_channels
self.decoder = TAEHV(self.latent_channels)
load_state_dict(self.decoder, load_torch_file(decoder_path), ignore_start="encoder")
@torch.inference_mode()
def forward(self, x: torch.Tensor) -> torch.Tensor:
z = self.decoder.decode(x)
return z.squeeze(1)
def download_model(model_path: os.PathLike, model_url: str):
if not os.path.exists(model_path):
os.makedirs(os.path.dirname(model_path), exist_ok=True)
print(f'Downloading TAESD Model to: "{model_path}"...')
torch.hub.download_url_to_file(model_url, model_path)
def decoder_model():
latent_format: "LatentFormat" = shared.sd_model.model_config.latent_format
model_name: str = latent_format.taesd_decoder_name
if model_name is None:
return None
else:
_video = model_name in ["taew2_1"]
model_name = model_name + ".pth"
loaded_model = sd_vae_taesd_models.get(model_name)
if loaded_model is None:
model_path = os.path.join(paths_internal.models_path, "VAE-taesd", model_name)
download_model(model_path, (URL_V if _video else URL) + model_name)
if not os.path.exists(model_path):
return None
loaded_model = (TAEHVDecoder if _video else TAESDDecoder)(model_path, latent_format.latent_channels)
loaded_model.eval()
loaded_model.to(devices.device, devices.dtype)
sd_vae_taesd_models[model_name] = loaded_model
return loaded_model
def encoder_model():
latent_format: "LatentFormat" = shared.sd_model.model_config.latent_format
model_name: str = latent_format.taesd_decoder_name
if model_name is None:
return None
elif model_name in ["taew2_1"]:
return None
else:
model_name = model_name.replace("decoder", "encoder") + ".pth"
loaded_model = sd_vae_taesd_models.get(model_name)
if loaded_model is None:
model_path = os.path.join(paths_internal.models_path, "VAE-taesd", model_name)
download_model(model_path, URL + model_name)
if not os.path.exists(model_path):
return None
loaded_model = TAESDEncoder(model_path, latent_format.latent_channels)
loaded_model.eval()
loaded_model.to(devices.device, devices.dtype)
sd_vae_taesd_models[model_name] = loaded_model
return loaded_model