holm/easydiffusion
1
0
Fork 0
forked from extern/easydiffusion
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Change the backend to a custom fork of SD, since basujindal's fork is no longer under development. This fork is intended to include the common models/tools used like RealESRGAN, GFPGAN, Codeformer etc, and is meant to be a community-developed project

Browse Source
This commit is contained in:
cmdr2 2022-11-22 16:38:39 +05:30
parent 040d7a6563
commit 93bbfac29a
6 changed files with 54 additions and 236 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
CHANGES.md
View File

@ -21,6 +21,7 @@
- A `What's New?` tab in the UI
### Detailed changelog
* 2.4.14 - 22 Nov 2022 - Change the backend to a custom fork of Stable Diffusion
* 2.4.13 - 21 Nov 2022 - Change the modifier weight via mouse wheel, drag to reorder selected modifiers, and some more modifier-related fixes. Thanks @patriceac
* 2.4.12 - 21 Nov 2022 - Another fix for improving how long images take to generate. Reduces the time taken for an enqueued task to start processing.
* 2.4.11 - 21 Nov 2022 - Installer improvements: avoid crashing if the username contains a space or special characters, allow moving/renaming the folder after installation on Windows, whitespace fix on git apply

10
scripts/on_sd_start.bat
View File

@ -33,18 +33,19 @@ if exist "Open Developer Console.cmd" del "Open Developer Console.cmd"
@cd stable-diffusion
@call git remote set-url origin https://github.com/easydiffusion/diffusion-kit.git
@call git reset --hard
@call git pull
@call git -c advice.detachedHead=false checkout f6cfebffa752ee11a7b07497b8529d5971de916c
@call git -c advice.detachedHead=false checkout 675fdf5c5694b3590f86583112f70794fa17052f
@call git apply --whitespace=nowarn ..\ui\sd_internal\ddim_callback.patch
@call git apply --whitespace=nowarn ..\ui\sd_internal\env_yaml.patch
@cd ..
) else (
@echo. & echo "Downloading Stable Diffusion.." & echo.
@call git clone https://github.com/basujindal/stable-diffusion.git && (
@call git clone https://github.com/easydiffusion/diffusion-kit.git stable-diffusion && (
@echo sd_git_cloned >> scripts\install_status.txt
) || (
@echo "Error downloading Stable Diffusion. Sorry about that, please try to:" & echo " 1. Run this installer again." & echo " 2. If that doesn't fix it, please try the common troubleshooting steps at https://github.com/cmdr2/stable-diffusion-ui/wiki/Troubleshooting" & echo " 3. If those steps don't help, please copy *all* the error messages in this window, and ask the community at https://discord.com/invite/u9yhsFmEkB" & echo " 4. If that doesn't solve the problem, please file an issue at https://github.com/cmdr2/stable-diffusion-ui/issues" & echo "Thanks!"
@ -53,10 +54,9 @@ if exist "Open Developer Console.cmd" del "Open Developer Console.cmd"
)
@cd stable-diffusion
@call git -c advice.detachedHead=false checkout f6cfebffa752ee11a7b07497b8529d5971de916c
@call git -c advice.detachedHead=false checkout 675fdf5c5694b3590f86583112f70794fa17052f
@call git apply --whitespace=nowarn ..\ui\sd_internal\ddim_callback.patch
@call git apply --whitespace=nowarn ..\ui\sd_internal\env_yaml.patch
@cd ..
)

10
scripts/on_sd_start.sh
View File

@ -26,28 +26,28 @@ if [ -e "scripts/install_status.txt" ] && [ `grep -c sd_git_cloned scripts/insta
cd stable-diffusion
git remote set-url origin https://github.com/easydiffusion/diffusion-kit.git
git reset --hard
git pull
git -c advice.detachedHead=false checkout f6cfebffa752ee11a7b07497b8529d5971de916c
git -c advice.detachedHead=false checkout 675fdf5c5694b3590f86583112f70794fa17052f
git apply --whitespace=nowarn ../ui/sd_internal/ddim_callback.patch || fail "ddim patch failed"
git apply --whitespace=nowarn ../ui/sd_internal/env_yaml.patch || fail "yaml patch failed"
cd ..
else
printf "\n\nDownloading Stable Diffusion..\n\n"
if git clone https://github.com/basujindal/stable-diffusion.git ; then
if git clone https://github.com/easydiffusion/diffusion-kit.git stable-diffusion ; then
echo sd_git_cloned >> scripts/install_status.txt
else
fail "git clone of basujindal/stable-diffusion.git failed"
fi
cd stable-diffusion
git -c advice.detachedHead=false checkout f6cfebffa752ee11a7b07497b8529d5971de916c
git -c advice.detachedHead=false checkout 675fdf5c5694b3590f86583112f70794fa17052f
git apply --whitespace=nowarn ../ui/sd_internal/ddim_callback.patch || fail "ddim patch failed"
git apply --whitespace=nowarn ../ui/sd_internal/env_yaml.patch || fail "yaml patch failed"
cd ..
fi

2
ui/index.html
View File

@ -22,7 +22,7 @@
<div id="logo">
<h1>
Stable Diffusion UI
<small>v2.4.13 <span id="updateBranchLabel"></span></small>
<small>v2.4.14 <span id="updateBranchLabel"></span></small>
</h1>
</div>
<div id="server-status">

254
ui/sd_internal/ddim_callback.patch
View File

@ -1,72 +1,13 @@
diff --git a/optimizedSD/ddpm.py b/optimizedSD/ddpm.py
index b967b55..35ef520 100644
index 79058bc..a473411 100644
--- a/optimizedSD/ddpm.py
+++ b/optimizedSD/ddpm.py
@@ -22,7 +22,7 @@ from ldm.util import exists, default, instantiate_from_config
from ldm.modules.diffusionmodules.util import make_beta_schedule
from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like
from ldm.modules.diffusionmodules.util import make_beta_schedule, extract_into_tensor, noise_like
-from samplers import CompVisDenoiser, get_ancestral_step, to_d, append_dims,linear_multistep_coeff
+from .samplers import CompVisDenoiser, get_ancestral_step, to_d, append_dims,linear_multistep_coeff
@@ -564,12 +564,12 @@ class UNet(DDPM):
unconditional_guidance_scale=unconditional_guidance_scale,
callback=callback, img_callback=img_callback)
def disabled_train(self):
"""Overwrite model.train with this function to make sure train/eval mode
@@ -506,6 +506,8 @@ class UNet(DDPM):
x_latent = noise if x0 is None else x0
# sampling
+ if sampler in ('ddim', 'dpm2', 'heun', 'dpm2_a', 'lms') and not hasattr(self, 'ddim_timesteps'):
+ self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=False)
if sampler == "plms":
self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=False)
@@ -528,39 +530,46 @@ class UNet(DDPM):
elif sampler == "ddim":
samples = self.ddim_sampling(x_latent, conditioning, S, unconditional_guidance_scale=unconditional_guidance_scale,
unconditional_conditioning=unconditional_conditioning,
- mask = mask,init_latent=x_T,use_original_steps=False)
+ mask = mask,init_latent=x_T,use_original_steps=False,
+ callback=callback, img_callback=img_callback)
elif sampler == "euler":
self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=False)
samples = self.euler_sampling(self.alphas_cumprod,x_latent, S, conditioning, unconditional_conditioning=unconditional_conditioning,
- unconditional_guidance_scale=unconditional_guidance_scale)
+ unconditional_guidance_scale=unconditional_guidance_scale,
+ img_callback=img_callback)
elif sampler == "euler_a":
self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=False)
samples = self.euler_ancestral_sampling(self.alphas_cumprod,x_latent, S, conditioning, unconditional_conditioning=unconditional_conditioning,
- unconditional_guidance_scale=unconditional_guidance_scale)
+ unconditional_guidance_scale=unconditional_guidance_scale,
+ img_callback=img_callback)
elif sampler == "dpm2":
samples = self.dpm_2_sampling(self.alphas_cumprod,x_latent, S, conditioning, unconditional_conditioning=unconditional_conditioning,
- unconditional_guidance_scale=unconditional_guidance_scale)
+ unconditional_guidance_scale=unconditional_guidance_scale,
+ img_callback=img_callback)
elif sampler == "heun":
samples = self.heun_sampling(self.alphas_cumprod,x_latent, S, conditioning, unconditional_conditioning=unconditional_conditioning,
- unconditional_guidance_scale=unconditional_guidance_scale)
+ unconditional_guidance_scale=unconditional_guidance_scale,
+ img_callback=img_callback)
elif sampler == "dpm2_a":
samples = self.dpm_2_ancestral_sampling(self.alphas_cumprod,x_latent, S, conditioning, unconditional_conditioning=unconditional_conditioning,
- unconditional_guidance_scale=unconditional_guidance_scale)
+ unconditional_guidance_scale=unconditional_guidance_scale,
+ img_callback=img_callback)
elif sampler == "lms":
samples = self.lms_sampling(self.alphas_cumprod,x_latent, S, conditioning, unconditional_conditioning=unconditional_conditioning,
- unconditional_guidance_scale=unconditional_guidance_scale)
+ unconditional_guidance_scale=unconditional_guidance_scale,
+ img_callback=img_callback)
+
+ yield from samples
+
if(self.turbo):
self.model1.to("cpu")
self.model2.to("cpu")
@ -76,7 +17,7 @@ index b967b55..35ef520 100644
@torch.no_grad()
def plms_sampling(self, cond,b, img,
ddim_use_original_steps=False,
@@ -599,10 +608,10 @@ class UNet(DDPM):
@@ -608,10 +608,10 @@ class UNet(DDPM):
old_eps.append(e_t)
if len(old_eps) >= 4:
old_eps.pop(0)
@ -90,23 +31,15 @@ index b967b55..35ef520 100644
@torch.no_grad()
def p_sample_plms(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
@@ -706,7 +715,8 @@ class UNet(DDPM):
@torch.no_grad()
def ddim_sampling(self, x_latent, cond, t_start, unconditional_guidance_scale=1.0, unconditional_conditioning=None,
- mask = None,init_latent=None,use_original_steps=False):
+ mask = None,init_latent=None,use_original_steps=False,
+ callback=None, img_callback=None):
timesteps = self.ddim_timesteps
timesteps = timesteps[:t_start]
@@ -730,10 +740,13 @@ class UNet(DDPM):
@@ -740,13 +740,13 @@ class UNet(DDPM):
unconditional_guidance_scale=unconditional_guidance_scale,
unconditional_conditioning=unconditional_conditioning)
- if callback: callback(i)
- if img_callback: img_callback(x_dec, i)
+ if callback: yield from callback(i)
+ if img_callback: yield from img_callback(x_dec, i)
+
if mask is not None:
- return x0 * mask + (1. - mask) * x_dec
+ x_dec = x0 * mask + (1. - mask) * x_dec
@ -116,217 +49,114 @@ index b967b55..35ef520 100644
@torch.no_grad()
@@ -779,13 +792,16 @@ class UNet(DDPM):
@@ -820,12 +820,12 @@ class UNet(DDPM):
@torch.no_grad()
- def euler_sampling(self, ac, x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1,extra_args=None,callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.):
+ def euler_sampling(self, ac, x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1,extra_args=None,callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.,
+ img_callback=None):
"""Implements Algorithm 2 (Euler steps) from Karras et al. (2022)."""
extra_args = {} if extra_args is None else extra_args
cvd = CompVisDenoiser(ac)
sigmas = cvd.get_sigmas(S)
x = x*sigmas[0]
+ print(f"Running Euler Sampling with {len(sigmas) - 1} timesteps")
+
s_in = x.new_ones([x.shape[0]]).half()
for i in trange(len(sigmas) - 1, disable=disable):
gamma = min(s_churn / (len(sigmas) - 1), 2 ** 0.5 - 1) if s_tmin <= sigmas[i] <= s_tmax else 0.
@@ -807,13 +823,18 @@ class UNet(DDPM):
d = to_d(x, sigma_hat, denoised)
if callback is not None:
callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigma_hat, 'denoised': denoised})
+
- if callback: callback(i)
- if img_callback: img_callback(x, i)
+ if callback: yield from callback(i)
+ if img_callback: yield from img_callback(x, i)
+
dt = sigmas[i + 1] - sigma_hat
# Euler method
x = x + d * dt
- return x
+
+ yield from img_callback(x, len(sigmas)-1)
@torch.no_grad()
- def euler_ancestral_sampling(self,ac,x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1,extra_args=None, callback=None, disable=None):
+ def euler_ancestral_sampling(self,ac,x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1,extra_args=None, callback=None, disable=None,
+ img_callback=None):
"""Ancestral sampling with Euler method steps."""
extra_args = {} if extra_args is None else extra_args
def euler_ancestral_sampling(self,ac,x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1,extra_args=None, callback=None, disable=None, img_callback=None):
@@ -852,14 +852,14 @@ class UNet(DDPM):
denoised = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
@@ -822,6 +843,8 @@ class UNet(DDPM):
sigmas = cvd.get_sigmas(S)
x = x*sigmas[0]
+ print(f"Running Euler Ancestral Sampling with {len(sigmas) - 1} timesteps")
+
s_in = x.new_ones([x.shape[0]]).half()
for i in trange(len(sigmas) - 1, disable=disable):
@@ -837,17 +860,22 @@ class UNet(DDPM):
sigma_down, sigma_up = get_ancestral_step(sigmas[i], sigmas[i + 1])
if callback is not None:
callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+
- if callback: callback(i)
- if img_callback: img_callback(x, i)
+ if callback: yield from callback(i)
+ if img_callback: yield from img_callback(x, i)
+
d = to_d(x, sigmas[i], denoised)
# Euler method
dt = sigma_down - sigmas[i]
x = x + d * dt
x = x + torch.randn_like(x) * sigma_up
- return x
+
+ yield from img_callback(x, len(sigmas)-1)
@torch.no_grad()
- def heun_sampling(self, ac, x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.):
+ def heun_sampling(self, ac, x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.,
+ img_callback=None):
"""Implements Algorithm 2 (Heun steps) from Karras et al. (2022)."""
extra_args = {} if extra_args is None else extra_args
@@ -892,8 +892,8 @@ class UNet(DDPM):
denoised = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
@@ -855,6 +883,8 @@ class UNet(DDPM):
sigmas = cvd.get_sigmas(S)
x = x*sigmas[0]
+ print(f"Running Heun Sampling with {len(sigmas) - 1} timesteps")
+
s_in = x.new_ones([x.shape[0]]).half()
for i in trange(len(sigmas) - 1, disable=disable):
@@ -876,6 +906,9 @@ class UNet(DDPM):
d = to_d(x, sigma_hat, denoised)
if callback is not None:
callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigma_hat, 'denoised': denoised})
+
- if callback: callback(i)
- if img_callback: img_callback(x, i)
+ if callback: yield from callback(i)
+ if img_callback: yield from img_callback(x, i)
+
dt = sigmas[i + 1] - sigma_hat
if sigmas[i + 1] == 0:
# Euler method
@@ -895,11 +928,13 @@ class UNet(DDPM):
@@ -913,7 +913,7 @@ class UNet(DDPM):
d_2 = to_d(x_2, sigmas[i + 1], denoised_2)
d_prime = (d + d_2) / 2
x = x + d_prime * dt
- return x
+
+ yield from img_callback(x, len(sigmas)-1)
@torch.no_grad()
- def dpm_2_sampling(self,ac,x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1,extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.):
+ def dpm_2_sampling(self,ac,x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1,extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.,
+ img_callback=None):
"""A sampler inspired by DPM-Solver-2 and Algorithm 2 from Karras et al. (2022)."""
extra_args = {} if extra_args is None else extra_args
@@ -907,6 +942,8 @@ class UNet(DDPM):
sigmas = cvd.get_sigmas(S)
x = x*sigmas[0]
+ print(f"Running DPM2 Sampling with {len(sigmas) - 1} timesteps")
+
s_in = x.new_ones([x.shape[0]]).half()
for i in trange(len(sigmas) - 1, disable=disable):
gamma = min(s_churn / (len(sigmas) - 1), 2 ** 0.5 - 1) if s_tmin <= sigmas[i] <= s_tmax else 0.
@@ -924,7 +961,7 @@ class UNet(DDPM):
@@ -944,8 +944,8 @@ class UNet(DDPM):
e_t_uncond, e_t = (x_in + eps * c_out).chunk(2)
denoised = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
-
- if callback: callback(i)
- if img_callback: img_callback(x, i)
+ if callback: yield from callback(i)
+ if img_callback: yield from img_callback(x, i)
d = to_d(x, sigma_hat, denoised)
# Midpoint method, where the midpoint is chosen according to a rho=3 Karras schedule
@@ -945,11 +982,13 @@ class UNet(DDPM):
@@ -966,7 +966,7 @@ class UNet(DDPM):
d_2 = to_d(x_2, sigma_mid, denoised_2)
x = x + d_2 * dt_2
- return x
+
+ yield from img_callback(x, len(sigmas)-1)
@torch.no_grad()
- def dpm_2_ancestral_sampling(self,ac,x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1, extra_args=None, callback=None, disable=None):
+ def dpm_2_ancestral_sampling(self,ac,x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1, extra_args=None, callback=None, disable=None,
+ img_callback=None):
"""Ancestral sampling with DPM-Solver inspired second-order steps."""
extra_args = {} if extra_args is None else extra_args
@@ -994,8 +994,8 @@ class UNet(DDPM):
@@ -957,6 +996,8 @@ class UNet(DDPM):
sigmas = cvd.get_sigmas(S)
x = x*sigmas[0]
+ print(f"Running DPM2 Ancestral Sampling with {len(sigmas) - 1} timesteps")
+
s_in = x.new_ones([x.shape[0]]).half()
for i in trange(len(sigmas) - 1, disable=disable):
@@ -973,6 +1014,9 @@ class UNet(DDPM):
sigma_down, sigma_up = get_ancestral_step(sigmas[i], sigmas[i + 1])
if callback is not None:
callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
+
- if callback: callback(i)
- if img_callback: img_callback(x, i)
+ if callback: yield from callback(i)
+ if img_callback: yield from img_callback(x, i)
+
d = to_d(x, sigmas[i], denoised)
# Midpoint method, where the midpoint is chosen according to a rho=3 Karras schedule
sigma_mid = ((sigmas[i] ** (1 / 3) + sigma_down ** (1 / 3)) / 2) ** 3
@@ -993,11 +1037,13 @@ class UNet(DDPM):
@@ -1016,7 +1016,7 @@ class UNet(DDPM):
d_2 = to_d(x_2, sigma_mid, denoised_2)
x = x + d_2 * dt_2
x = x + torch.randn_like(x) * sigma_up
- return x
+
+ yield from img_callback(x, len(sigmas)-1)
@torch.no_grad()
- def lms_sampling(self,ac,x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1, extra_args=None, callback=None, disable=None, order=4):
+ def lms_sampling(self,ac,x, S, cond, unconditional_conditioning = None, unconditional_guidance_scale = 1, extra_args=None, callback=None, disable=None, order=4,
+ img_callback=None):
extra_args = {} if extra_args is None else extra_args
s_in = x.new_ones([x.shape[0]])
@@ -1005,6 +1051,8 @@ class UNet(DDPM):
sigmas = cvd.get_sigmas(S)
x = x*sigmas[0]
+ print(f"Running LMS Sampling with {len(sigmas) - 1} timesteps")
+
ds = []
for i in trange(len(sigmas) - 1, disable=disable):
@@ -1017,6 +1065,7 @@ class UNet(DDPM):
@@ -1042,8 +1042,8 @@ class UNet(DDPM):
e_t_uncond, e_t = (x_in + eps * c_out).chunk(2)
denoised = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
- if callback: callback(i)
- if img_callback: img_callback(x, i)
+ if callback: yield from callback(i)
+ if img_callback: yield from img_callback(x, i)
d = to_d(x, sigmas[i], denoised)
ds.append(d)
@@ -1027,4 +1076,5 @@ class UNet(DDPM):
@@ -1054,4 +1054,4 @@ class UNet(DDPM):
cur_order = min(i + 1, order)
coeffs = [linear_multistep_coeff(cur_order, sigmas.cpu(), i, j) for j in range(cur_order)]
x = x + sum(coeff * d for coeff, d in zip(coeffs, reversed(ds)))
- return x
+
+ yield from img_callback(x, len(sigmas)-1)
diff --git a/optimizedSD/openaimodelSplit.py b/optimizedSD/openaimodelSplit.py
index abc3098..7a32ffe 100644
--- a/optimizedSD/openaimodelSplit.py
+++ b/optimizedSD/openaimodelSplit.py
@@ -13,7 +13,7 @@ from ldm.modules.diffusionmodules.util import (
normalization,
timestep_embedding,
)
-from splitAttention import SpatialTransformer
+from .splitAttention import SpatialTransformer
class AttentionPool2d(nn.Module):

13
ui/sd_internal/env_yaml.patch
View File

@ -1,13 +0,0 @@
diff --git a/environment.yaml b/environment.yaml
index 7f25da8..306750f 100644
--- a/environment.yaml
+++ b/environment.yaml
@@ -23,6 +23,8 @@ dependencies:
- torch-fidelity==0.3.0
- transformers==4.19.2
- torchmetrics==0.6.0
+ - pywavelets==1.3.0
+ - pandas==1.4.4
- kornia==0.6
- -e git+https://github.com/CompVis/taming-transformers.git@master#egg=taming-transformers
- -e git+https://github.com/openai/CLIP.git@main#egg=clip