This commit is contained in:
AUTOMATIC 2022-08-22 17:15:46 +03:00
parent 71cf18b012
commit 3324f31e84
3 changed files with 462 additions and 2 deletions

View file

@ -1,2 +1,58 @@
# stable-diffusion-webui
Stable Diffusion web UI
# Stable Diffusion web UI
A browser interface based on Gradio library for Stable Diffusion.
Original script with Gradio UI was written by a kind anonymopus user. This is a modification.
![](screenshot.png)
## Stable Diffusion
This script assumes that you already have main Stable Diffusion sutff installed, assumed to be in directory `/sd`.
If you don't have it installed, follow the guide:
- https://rentry.org/kretard
This repository's `webgui.py` is a replacement for `kdiff.py` from the guide.
Particularly, following files must exist:
- `/sd/configs/stable-diffusion/v1-inference.yaml`
- `/sd/models/ldm/stable-diffusion-v1/model.ckpt`
- `/sd/ldm/util.py`
- `/sd/k_diffusion/__init__.py`
## GFPGAN
If you want to use GFPGAN to improve generated faces, you need to install it separately.
Follow instructions from https://github.com/TencentARC/GFPGAN, but when cloning it, do so into Stable Diffusion main directory, `/sd`.
After that download [GFPGANv1.3.pth](https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth) and put it
into the `/sd/GFPGAN/experiments/pretrained_models` directory. If you're getting troubles with GFPGAN support, follow instructions
from the GFPGAN's repository until `inference_gfpgan.py` script works.
The following files must exist:
- `/sd/GFPGAN/inference_gfpgan.py`
- `/sd/GFPGAN/experiments/pretrained_models/GFPGANv1.3.pth`
If the GFPGAN directory does not exist, you will not get the option to use GFPGAN in the UI. If it does exist, you will either be able
to use it, or there will be a message in console with an error related to GFPGAN.
## Web UI
Run the script as:
`python webui.py`
When running the script, you must be in the main Stable Diffusion directory, `/sd`. If you cloned this repository into a subdirectory
of `/sd`, say, the `stable-diffusion-webui` directory, you will run it as:
`python stable-diffusion-webui/webui.py`
When launching, you may get a very long warning message related to some weights not being used. You may freely ignore it.
After a while, you will get a message like this:
```
Running on local URL: http://127.0.0.1:7860/
```
Open the URL in browser, and you are good to go.

BIN
screenshot.png Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 865 KiB

404
webui.py Normal file
View file

@ -0,0 +1,404 @@
import PIL
import argparse, os, sys, glob
import torch
import torch.nn as nn
import numpy as np
import gradio as gr
from omegaconf import OmegaConf
from PIL import Image
from itertools import islice
from einops import rearrange, repeat
from torchvision.utils import make_grid
from torch import autocast
from contextlib import contextmanager, nullcontext
import mimetypes
import random
import k_diffusion as K
from ldm.util import instantiate_from_config
from ldm.models.diffusion.ddim import DDIMSampler
from ldm.models.diffusion.plms import PLMSSampler
# this is a fix for Windows users. Without it, javascript files will be served with text/html content-type and the bowser will not show any UI
mimetypes.init()
mimetypes.add_type('application/javascript', '.js')
# some of those options should not be changed at all because they would break the model, so I removed them from options.
opt_C = 4
opt_f = 8
parser = argparse.ArgumentParser()
parser.add_argument("--outdir", type=str, nargs="?", help="dir to write results to", default=None)
parser.add_argument("--skip_grid", action='store_true', help="do not save a grid, only individual samples. Helpful when evaluating lots of samples",)
parser.add_argument("--skip_save", action='store_true', help="do not save indiviual samples. For speed measurements.",)
parser.add_argument("--n_rows", type=int, default=0, help="rows in the grid (default: n_samples)",)
parser.add_argument("--config", type=str, default="configs/stable-diffusion/v1-inference.yaml", help="path to config which constructs model",)
parser.add_argument("--ckpt", type=str, default="models/ldm/stable-diffusion-v1/model.ckpt", help="path to checkpoint of model",)
parser.add_argument("--precision", type=str, help="evaluate at this precision", choices=["full", "autocast"], default="autocast")
parser.add_argument("--gfpgan-dir", type=str, help="GFPGAN directory", default='./GFPGAN')
opt = parser.parse_args()
GFPGAN_dir = opt.gfpgan_dir
def chunk(it, size):
it = iter(it)
return iter(lambda: tuple(islice(it, size)), ())
def load_model_from_config(config, ckpt, verbose=False):
print(f"Loading model from {ckpt}")
pl_sd = torch.load(ckpt, map_location="cpu")
if "global_step" in pl_sd:
print(f"Global Step: {pl_sd['global_step']}")
sd = pl_sd["state_dict"]
model = instantiate_from_config(config.model)
m, u = model.load_state_dict(sd, strict=False)
if len(m) > 0 and verbose:
print("missing keys:")
print(m)
if len(u) > 0 and verbose:
print("unexpected keys:")
print(u)
model.cuda()
model.eval()
return model
def load_img_pil(img_pil):
image = img_pil.convert("RGB")
w, h = image.size
print(f"loaded input image of size ({w}, {h})")
w, h = map(lambda x: x - x % 64, (w, h)) # resize to integer multiple of 64
image = image.resize((w, h), resample=PIL.Image.LANCZOS)
print(f"cropped image to size ({w}, {h})")
image = np.array(image).astype(np.float32) / 255.0
image = image[None].transpose(0, 3, 1, 2)
image = torch.from_numpy(image)
return 2. * image - 1.
def load_img(path):
return load_img_pil(Image.open(path))
class CFGDenoiser(nn.Module):
def __init__(self, model):
super().__init__()
self.inner_model = model
def forward(self, x, sigma, uncond, cond, cond_scale):
x_in = torch.cat([x] * 2)
sigma_in = torch.cat([sigma] * 2)
cond_in = torch.cat([uncond, cond])
uncond, cond = self.inner_model(x_in, sigma_in, cond=cond_in).chunk(2)
return uncond + (cond - uncond) * cond_scale
def load_GFPGAN():
model_name = 'GFPGANv1.3'
model_path = os.path.join(GFPGAN_dir, 'experiments/pretrained_models', model_name + '.pth')
if not os.path.isfile(model_path):
raise Exception("GFPGAN model not found at path "+model_path)
sys.path.append(os.path.abspath(GFPGAN_dir))
from gfpgan import GFPGANer
return GFPGANer(model_path=model_path, upscale=1, arch='clean', channel_multiplier=2, bg_upsampler=None)
GFPGAN = None
if os.path.exists(GFPGAN_dir):
try:
GFPGAN = load_GFPGAN()
print("Loaded GFPGAN")
except Exception:
import traceback
print("Error loading GFPGAN:", file=sys.stderr)
print(traceback.format_exc(), file=sys.stderr)
config = OmegaConf.load("configs/stable-diffusion/v1-inference.yaml")
model = load_model_from_config(config, "models/ldm/stable-diffusion-v1/model.ckpt")
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
model = model.half().to(device)
def image_grid(imgs, rows):
cols = len(imgs) // rows
w, h = imgs[0].size
grid = Image.new('RGB', size=(cols * w, rows * h))
for i, img in enumerate(imgs):
grid.paste(img, box=(i % cols * w, i // cols * h))
return grid
def dream(prompt: str, ddim_steps: int, sampler_name: str, fixed_code: bool, use_GFPGAN: bool, ddim_eta: float, n_iter: int, n_samples: int, cfg_scale: float, seed: int, height: int, width: int):
torch.cuda.empty_cache()
outpath = opt.outdir or "outputs/txt2img-samples"
if seed == -1:
seed = random.randrange(4294967294)
seed = int(seed)
is_PLMS = sampler_name == 'PLMS'
is_DDIM = sampler_name == 'DDIM'
is_Kdif = sampler_name == 'k-diffusion'
sampler = None
if is_PLMS:
sampler = PLMSSampler(model)
elif is_DDIM:
sampler = DDIMSampler(model)
elif is_Kdif:
pass
else:
raise Exception("Unknown sampler: " + sampler_name)
model_wrap = K.external.CompVisDenoiser(model)
os.makedirs(outpath, exist_ok=True)
batch_size = n_samples
n_rows = opt.n_rows if opt.n_rows > 0 else batch_size
assert prompt is not None
data = [batch_size * [prompt]]
sample_path = os.path.join(outpath, "samples")
os.makedirs(sample_path, exist_ok=True)
base_count = len(os.listdir(sample_path))
grid_count = len(os.listdir(outpath)) - 1
start_code = None
if fixed_code:
start_code = torch.randn([n_samples, opt_C, height // opt_f, width // opt_f], device=device)
precision_scope = autocast if opt.precision == "autocast" else nullcontext
output_images = []
with torch.no_grad(), precision_scope("cuda"), model.ema_scope():
all_samples = []
for n in range(n_iter):
for batch_index, prompts in enumerate(data):
uc = None
if cfg_scale != 1.0:
uc = model.get_learned_conditioning(batch_size * [""])
if isinstance(prompts, tuple):
prompts = list(prompts)
c = model.get_learned_conditioning(prompts)
shape = [opt_C, height // opt_f, width // opt_f]
current_seed = seed + n * len(data) + batch_index
torch.manual_seed(current_seed)
if is_Kdif:
sigmas = model_wrap.get_sigmas(ddim_steps)
x = torch.randn([n_samples, *shape], device=device) * sigmas[0] # for GPU draw
model_wrap_cfg = CFGDenoiser(model_wrap)
samples_ddim = K.sampling.sample_lms(model_wrap_cfg, x, sigmas, extra_args={'cond': c, 'uncond': uc, 'cond_scale': cfg_scale}, disable=False)
elif sampler is not None:
samples_ddim, _ = sampler.sample(S=ddim_steps, conditioning=c, batch_size=n_samples, shape=shape, verbose=False, unconditional_guidance_scale=cfg_scale, unconditional_conditioning=uc, eta=ddim_eta, x_T=start_code)
x_samples_ddim = model.decode_first_stage(samples_ddim)
x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
if not opt.skip_save or not opt.skip_grid:
for x_sample in x_samples_ddim:
x_sample = 255. * rearrange(x_sample.cpu().numpy(), 'c h w -> h w c')
x_sample = x_sample.astype(np.uint8)
if use_GFPGAN and GFPGAN is not None:
cropped_faces, restored_faces, restored_img = GFPGAN.enhance(x_sample, has_aligned=False, only_center_face=False, paste_back=True)
x_sample = restored_img
image = Image.fromarray(x_sample)
image.save(os.path.join(sample_path, f"{base_count:05}-{current_seed}_{prompt.replace(' ', '_')[:128]}.png"))
output_images.append(image)
base_count += 1
if not opt.skip_grid:
all_samples.append(x_sample)
if not opt.skip_grid:
# additionally, save as grid
grid = image_grid(output_images, rows=n_rows)
grid.save(os.path.join(outpath, f'grid-{grid_count:04}.png'))
grid_count += 1
if sampler is not None:
del sampler
info = f"""
{prompt}
Steps: {ddim_steps}, Sampler: {sampler_name}, CFG scale: {cfg_scale}, Seed: {seed}{', GFPGAN' if use_GFPGAN and GFPGAN is not None else ''}
""".strip()
return output_images, seed, info
dream_interface = gr.Interface(
dream,
inputs=[
gr.Textbox(label="Prompt", placeholder="A corgi wearing a top hat as an oil painting.", lines=1),
gr.Slider(minimum=1, maximum=150, step=1, label="Sampling Steps", value=50),
gr.Radio(label='Sampling method', choices=["DDIM", "PLMS", "k-diffusion"], value="k-diffusion"),
gr.Checkbox(label='Enable Fixed Code sampling', value=False),
gr.Checkbox(label='Fix faces using GFPGAN', value=False, visible=GFPGAN is not None),
gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label="DDIM ETA", value=0.0, visible=False),
gr.Slider(minimum=1, maximum=16, step=1, label='Sampling iterations', value=1),
gr.Slider(minimum=1, maximum=4, step=1, label='Samples per iteration', value=1),
gr.Slider(minimum=1.0, maximum=15.0, step=0.5, label='Classifier Free Guidance Scale', value=7.0),
gr.Number(label='Seed', value=-1),
gr.Slider(minimum=64, maximum=2048, step=64, label="Height", value=512),
gr.Slider(minimum=64, maximum=2048, step=64, label="Width", value=512),
],
outputs=[
gr.Gallery(label="Images"),
gr.Number(label='Seed'),
gr.Textbox(label="Copy-paste generation parameters"),
],
title="Stable Diffusion Text-to-Image K",
description="Generate images from text with Stable Diffusion (using K-LMS)",
allow_flagging="never"
)
def translation(prompt: str, init_img, ddim_steps: int, ddim_eta: float, n_iter: int, n_samples: int, cfg_scale: float, denoising_strength: float, seed: int, height: int, width: int):
torch.cuda.empty_cache()
outpath = opt.outdir or "outputs/img2img-samples"
if seed == -1:
seed = random.randrange(4294967294)
sampler = DDIMSampler(model)
model_wrap = K.external.CompVisDenoiser(model)
os.makedirs(outpath, exist_ok=True)
batch_size = n_samples
n_rows = opt.n_rows if opt.n_rows > 0 else batch_size
assert prompt is not None
data = [batch_size * [prompt]]
sample_path = os.path.join(outpath, "samples")
os.makedirs(sample_path, exist_ok=True)
base_count = len(os.listdir(sample_path))
grid_count = len(os.listdir(outpath)) - 1
seedit = 0
image = init_img.convert("RGB")
w, h = image.size
image = np.array(image).astype(np.float32) / 255.0
image = image[None].transpose(0, 3, 1, 2)
image = torch.from_numpy(image)
output_images = []
precision_scope = autocast if opt.precision == "autocast" else nullcontext
with torch.no_grad():
with precision_scope("cuda"):
init_image = 2. * image - 1.
init_image = init_image.to(device)
init_image = repeat(init_image, '1 ... -> b ...', b=batch_size)
init_latent = model.get_first_stage_encoding(model.encode_first_stage(init_image)) # move to latent space
x0 = init_latent
sampler.make_schedule(ddim_num_steps=ddim_steps, ddim_eta=ddim_eta, verbose=False)
assert 0. <= denoising_strength <= 1., 'can only work with strength in [0.0, 1.0]'
t_enc = int(denoising_strength * ddim_steps)
print(f"target t_enc is {t_enc} steps")
with model.ema_scope():
all_samples = list()
for n in range(n_iter):
for batch_index, prompts in enumerate(data):
uc = None
if cfg_scale != 1.0:
uc = model.get_learned_conditioning(batch_size * [""])
if isinstance(prompts, tuple):
prompts = list(prompts)
c = model.get_learned_conditioning(prompts)
sigmas = model_wrap.get_sigmas(ddim_steps)
current_seed = seed + n * len(data) + batch_index
torch.manual_seed(current_seed)
noise = torch.randn_like(x0) * sigmas[ddim_steps - t_enc - 1] # for GPU draw
xi = x0 + noise
sigma_sched = sigmas[ddim_steps - t_enc - 1:]
# x = torch.randn([n_samples, *shape]).to(device) * sigmas[0] # for CPU draw
model_wrap_cfg = CFGDenoiser(model_wrap)
extra_args = {'cond': c, 'uncond': uc, 'cond_scale': cfg_scale}
samples_ddim = K.sampling.sample_lms(model_wrap_cfg, xi, sigma_sched, extra_args=extra_args, disable=False)
x_samples_ddim = model.decode_first_stage(samples_ddim)
x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
if not opt.skip_save:
for x_sample in x_samples_ddim:
x_sample = 255. * rearrange(x_sample.cpu().numpy(), 'c h w -> h w c')
image = Image.fromarray(x_sample.astype(np.uint8))
image.save(os.path.join(sample_path, f"{base_count:05}-{current_seed}_{prompt.replace(' ', '_')[:128]}.png"))
output_images.append(image)
base_count += 1
seedit += 1
if not opt.skip_grid:
all_samples.append(x_samples_ddim)
if not opt.skip_grid:
# additionally, save as grid
grid = torch.stack(all_samples, 0)
grid = rearrange(grid, 'n b c h w -> (n b) c h w')
grid = make_grid(grid, nrow=n_rows)
# to image
grid = 255. * rearrange(grid, 'c h w -> h w c').cpu().numpy()
Image.fromarray(grid.astype(np.uint8)).save(os.path.join(outpath, f'grid-{grid_count:04}.png'))
Image.fromarray(grid.astype(np.uint8))
grid_count += 1
del sampler
return output_images, seed
# prompt, init_img, ddim_steps, plms, ddim_eta, n_iter, n_samples, cfg_scale, denoising_strength, seed
img2img_interface = gr.Interface(
translation,
inputs=[
gr.Textbox(placeholder="A fantasy landscape, trending on artstation.", lines=1),
gr.Image(value="https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg", source="upload", interactive=True, type="pil"),
gr.Slider(minimum=1, maximum=150, step=1, label="Sampling Steps", value=50),
gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label="DDIM ETA", value=0.0, visible=False),
gr.Slider(minimum=1, maximum=50, step=1, label='Sampling iterations', value=2),
gr.Slider(minimum=1, maximum=8, step=1, label='Samples per iteration', value=2),
gr.Slider(minimum=1.0, maximum=15.0, step=0.5, label='Classifier Free Guidance Scale', value=7.0),
gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising Strength', value=0.75),
gr.Number(label='Seed', value=-1),
gr.Slider(minimum=64, maximum=2048, step=64, label="Resize Height", value=512),
gr.Slider(minimum=64, maximum=2048, step=64, label="Resize Width", value=512),
],
outputs=[
gr.Gallery(),
gr.Number(label='Seed')
],
title="Stable Diffusion Image-to-Image",
description="Generate images from images with Stable Diffusion",
)
demo = gr.TabbedInterface(interface_list=[dream_interface, img2img_interface], tab_names=["Dream", "Image Translation"])
demo.launch()