317 lines
9.2 KiB
Python
317 lines
9.2 KiB
Python
# forked from https://github.com/madebyollin/taesd/blob/main/taesd.py
|
|
|
|
"""
|
|
Tiny AutoEncoder for Stable Diffusion
|
|
(DNN for encoding / decoding SD's latent space)
|
|
"""
|
|
|
|
import torch
|
|
import torch.nn as nn
|
|
|
|
|
|
def conv(n_in, n_out, **kwargs):
|
|
return nn.Conv2d(n_in, n_out, 3, padding=1, **kwargs)
|
|
|
|
|
|
class Clamp(nn.Module):
|
|
def forward(self, 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, n_gn = (
|
|
lambda n_in, n_out: nn.Conv2d(n_in, n_out, 1, bias=False),
|
|
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 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),
|
|
)
|
|
|
|
|
|
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 F32Encoder(latent_channels=32):
|
|
"""Encoder variant with 32x spatial downscaling instead of 8x."""
|
|
return nn.Sequential(
|
|
conv(3, 32, stride=2),
|
|
nn.ReLU(inplace=True),
|
|
conv(32, 64, stride=2),
|
|
nn.ReLU(inplace=True),
|
|
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),
|
|
Block(64, 64),
|
|
Block(64, 64),
|
|
conv(64, latent_channels),
|
|
)
|
|
|
|
|
|
def F32Decoder(latent_channels=32):
|
|
"""Decoder variant with 32x spatial upscaling instead of 8x."""
|
|
return nn.Sequential(
|
|
Clamp(),
|
|
conv(latent_channels, 256),
|
|
nn.ReLU(),
|
|
Block(256, 256),
|
|
Block(256, 256),
|
|
Block(256, 256),
|
|
nn.Upsample(scale_factor=2),
|
|
conv(256, 128, bias=False),
|
|
Block(128, 128),
|
|
Block(128, 128),
|
|
Block(128, 128),
|
|
nn.Upsample(scale_factor=2),
|
|
conv(128, 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),
|
|
Block(64, 64),
|
|
Block(64, 64),
|
|
nn.Upsample(scale_factor=2),
|
|
conv(64, 64, bias=False),
|
|
Block(64, 64),
|
|
conv(64, 3),
|
|
)
|
|
|
|
|
|
class TAESD(nn.Module):
|
|
latent_magnitude = 3
|
|
latent_shift = 0.5
|
|
|
|
def __init__(
|
|
self,
|
|
encoder_path="taesd_encoder.pth",
|
|
decoder_path="taesd_decoder.pth",
|
|
latent_channels=None,
|
|
arch_variant=None,
|
|
):
|
|
"""Initialize pretrained TAESD on the given device from the given checkpoints."""
|
|
super().__init__()
|
|
if latent_channels is None:
|
|
latent_channels, arch_variant = self.guess_latent_channels_and_arch(
|
|
str(encoder_path)
|
|
)
|
|
# flux_2 required global pooling/norm for accurate distillation, enable conditionally
|
|
self.encoder = Encoder(
|
|
latent_channels, use_midblock_gn=(arch_variant in ["flux_2"])
|
|
)
|
|
self.decoder = Decoder(
|
|
latent_channels, use_midblock_gn=(arch_variant in ["flux_2"])
|
|
)
|
|
# sana dcae requires 32x spatial downscaling, enable conditionally
|
|
if arch_variant == "f32":
|
|
self.encoder, self.decoder = (
|
|
F32Encoder(latent_channels),
|
|
F32Decoder(latent_channels),
|
|
)
|
|
if encoder_path is not None:
|
|
self.encoder.load_state_dict(
|
|
torch.load(encoder_path, map_location="cpu", weights_only=True)
|
|
)
|
|
if decoder_path is not None:
|
|
self.decoder.load_state_dict(
|
|
torch.load(decoder_path, map_location="cpu", weights_only=True)
|
|
)
|
|
|
|
def guess_latent_channels(self, encoder_path):
|
|
"""Guess latent channel count based on encoder filename"""
|
|
return self.guess_latent_channels_and_arch(encoder_path)[0]
|
|
|
|
def guess_latent_channels_and_arch(self, encoder_path):
|
|
"""Guess latent channel count and architecture variant based on encoder filename"""
|
|
if "taef1" in encoder_path:
|
|
return 16, None
|
|
if "taef2" in encoder_path:
|
|
return 32, "flux_2"
|
|
if "taesd3" in encoder_path:
|
|
return 16, None
|
|
if "taesana" in encoder_path:
|
|
return 32, "f32" # f32c32
|
|
return 4, None
|
|
|
|
@staticmethod
|
|
def scale_latents(x):
|
|
"""raw latents -> [0, 1]"""
|
|
return x.div(2 * TAESD.latent_magnitude).add(TAESD.latent_shift).clamp(0, 1)
|
|
|
|
@staticmethod
|
|
def unscale_latents(x):
|
|
"""[0, 1] -> raw latents"""
|
|
return x.sub(TAESD.latent_shift).mul(2 * TAESD.latent_magnitude)
|
|
|
|
|
|
@torch.no_grad()
|
|
def main():
|
|
from PIL import Image
|
|
import sys
|
|
import torchvision.transforms.functional as TF
|
|
|
|
dev = torch.device(
|
|
"cuda"
|
|
if torch.cuda.is_available()
|
|
else "mps"
|
|
if torch.backends.mps.is_available()
|
|
else "cpu"
|
|
)
|
|
print("Using device", dev)
|
|
taesd = TAESD().to(dev)
|
|
for im_path in sys.argv[1:]:
|
|
im = TF.to_tensor(Image.open(im_path).convert("RGB")).unsqueeze(0).to(dev)
|
|
|
|
# encode image, quantize, and save to file
|
|
im_enc = taesd.scale_latents(taesd.encoder(im)).mul_(255).round_().byte()
|
|
enc_path = im_path + ".encoded.png"
|
|
TF.to_pil_image(im_enc[0]).save(enc_path)
|
|
print(f"Encoded {im_path} to {enc_path}")
|
|
|
|
# load the saved file, dequantize, and decode
|
|
im_enc = taesd.unscale_latents(
|
|
TF.to_tensor(Image.open(enc_path)).unsqueeze(0).to(dev)
|
|
)
|
|
im_dec = taesd.decoder(im_enc).clamp(0, 1)
|
|
dec_path = im_path + ".decoded.png"
|
|
print(f"Decoded {enc_path} to {dec_path}")
|
|
TF.to_pil_image(im_dec[0]).save(dec_path)
|
|
|
|
|
|
# Construction
|
|
import torch
|
|
import safetensors.torch as stt
|
|
from diffusers.utils.accelerate_utils import apply_forward_hook
|
|
|
|
|
|
def convert_diffusers_sd_to_taesd(sd):
|
|
out = {}
|
|
for k, v in sd.items():
|
|
encdec, _layers, index, *suffix = k.split(".")
|
|
offset = 0
|
|
if encdec == "decoder":
|
|
offset = +1
|
|
out[".".join([encdec, str(int(index) + offset), *suffix])] = v
|
|
return out
|
|
|
|
|
|
class DotDict(dict):
|
|
__getattr__ = dict.__getitem__
|
|
__setattr__ = dict.__setitem__
|
|
|
|
|
|
class DiffusersTAEF2Wrapper(torch.nn.Module):
|
|
def __init__(self, path):
|
|
super().__init__()
|
|
self.dtype = torch.bfloat16
|
|
self.taesd = TAESD(
|
|
encoder_path=None,
|
|
decoder_path=None,
|
|
latent_channels=32,
|
|
arch_variant="flux_2",
|
|
).to(self.dtype)
|
|
self.taesd.load_state_dict(convert_diffusers_sd_to_taesd(stt.load_file(path)))
|
|
self.bn = torch.nn.BatchNorm2d(128, affine=False, eps=0.0) # default bn
|
|
self.config = DotDict(batch_norm_eps=self.bn.eps)
|
|
|
|
@apply_forward_hook
|
|
def encode(self, x):
|
|
return DotDict(
|
|
latent_dist=DotDict(
|
|
sample=lambda: self.taesd.encoder(
|
|
x.to(self.dtype).mul(0.5).add_(0.5)
|
|
).to(x.dtype)
|
|
)
|
|
)
|
|
|
|
@apply_forward_hook
|
|
def decode(self, x, return_dict=True):
|
|
x = (
|
|
self.taesd.decoder(x.to(self.dtype))
|
|
.mul(2)
|
|
.sub_(1)
|
|
.clamp_(-1, 1)
|
|
.to(x.dtype)
|
|
)
|
|
return (dict(sample=x) if return_dict else x,)
|