Skip to content

Instantly share code, notes, and snippets.

@esnya

esnya/easy_anime_face_expressions.py

Created July 17, 2023 17:59
Show Gist options
  • Save esnya/9276c87a37e72fc013385693b5e42d19 to your computer and use it in GitHub Desktop.
Save esnya/9276c87a37e72fc013385693b5e42d19 to your computer and use it in GitHub Desktop.
Download ZIP
Custom script for AUTOMATIC1111/stable-diffusion-webui to generate face expressions of anime characters. Powered by cascade classifier from nagadomi/lbpcascade_animeface.
Raw
easy_anime_face_expressions.py
"""
Easy Anime Face Expressions
Custom script for AUTOMATIC1111/stable-diffusion-webui to generate face expressions of anime characters.
Powered by cascade classifier from nagadomi/lbpcascade_animeface.
Usage:
1. Download this file to `scripts/easy_anime_face_expressions.py`.
2. Start the webui.
3. Generate base image with "closed mouth" in prompt. (or first expression tag you will put into the textbox.)
4. Press to "Send to img2img" to send the image and the prompt to img2img.
5. Select "img2img" tab.
6. Select "Easy Face Expressions" script from the dropdown menu "Script".
7. Press "Load defaults" button to load default values. (or fill the values manually.)
9. Press "Generate" button. And wait for the result.
License:
MIT License
Copyright (c) 2023 esnya
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
"""
from pathlib import Path
from typing import Any, Callable, cast
from httpx import request
import modules.scripts as scripts
from modules.processing import (
fix_seed,
process_images,
StableDiffusionProcessingImg2Img,
)
from modules.images import (
save_image,
draw_grid_annotations,
image_grid,
GridAnnotation,
)
import gradio as gr
import cv2
from PIL import Image
import numpy as np
from modules.shared import opts
default_expressions = ", ".join(
{
tag: tag
for tag in [
"closed mouth",
# tag group:face tags
"angry",
"anger vein",
"annoyed",
"clenched teeth",
"scowl",
"annoyed",
"blush",
"blush stickers",
"embarrassed",
"full-face blush",
"nose blush",
"bored",
"closed eyes",
"confused",
"crazy",
"desperation",
"determined",
"disappointed",
"disdain",
"disgust",
"despair",
"drunk",
"envy",
"evil",
"excited",
"exhausted",
"expressionless",
# "facepalm",
"flustered",
"frustrated",
"furrowed brow",
"grimace",
"guilt",
"happy",
"kubrick stare",
"lonely",
"nervous",
"nosebleed",
"one eye closed",
"open mouth",
"parted lips",
"pain",
"pout",
"raised eyebrow",
"rolling eyes",
"sad",
"depressed",
"frown",
"gloom",
"tears",
"scared",
"panicking",
"worried",
"serious",
"sigh",
"sleepy",
"sulking",
"surprised",
"thinking",
"pensive",
"v-shaped eyebrows",
"wince",
"upset",
"crazy smile",
"evil smile",
# "fingersmile",
"forced smile",
"glasgow smile",
"grin",
"evil grin",
"light smile",
"sad smile",
"seductive smile",
"stifled laugh",
"doyagao",
"smirk",
"smug",
"troll face",
"^^^",
"color drain",
"depressed",
"despair",
"gloom",
"horrified",
"screaming",
"sobbing",
"traumatized",
"turn pale",
"wavy mouth",
"upset",
# ";)",
# ":d",
# ";d",
# "xd",
# "d:",
# ":3",
# ";3",
# "x3",
# "3:",
# "uwu",
# ":p",
# ";p",
# ":q",
# ";q",
# ">:)",
# ">:(",
# ":t",
# ":i",
# ":/",
# ":|",
# ":x",
# ":c",
# "c:",
# ":<",
# ";<",
"diamond mouth",
# ":>",
# ":>=",
# ":o",
# ";o",
# "o3o",
# ">3<",
# "o_o",
# "0_0",
# "|_|",
# "._.",
"solid circle eyes",
"heart-shaped eyes",
# "^_^",
# "^o^",
# "\\(^o^)/",
# "^q^",
# ">_<",
# "xd",
# "x3",
# ">o<",
# "@_@",
# ">_@",
# "+_+",
# "+_-",
# "=_=",
# "=^=",
# "=v=",
# "<o>_<o>",
# "<|>_<|>",
"chestnut mouth",
"constricted pupils",
"cross-eyed",
"dot mouth",
"dot nose",
"mouth drool",
"no nose",
"no mouth",
"rectangular mouth",
"sideways mouth",
"split mouth",
"wavy mouth",
"wide-eyed",
"awesome face",
# "face of the people who sank all their money into the fx",
"foodgasm",
"henohenomoheji",
"nonowa",
"portrait",
"profile",
"smiley face",
"troll face",
"uso da",
"breast awe",
"food awe",
"muscle awe",
"staring",
]
}.values()
)
default_face_classifier = "lbpcascade_animeface"
default_min_neighbors = 9
default_scale_factor = 1.01
default_min_size = 128
default_mask_scaling = 0.9
default_mask_blur_size = 255
def load_defaults_fn():
return [
default_expressions,
default_face_classifier,
default_min_neighbors,
default_scale_factor,
default_min_size,
default_mask_scaling,
default_mask_blur_size,
]
def opts_setter(key: str) -> Callable[[Any], None]:
def setter(value):
opts.set(key, value)
return setter
def generate_face_mask(
image: Image.Image,
classifier: str,
min_neighbors: int,
scale_factor: float,
min_size: int,
mask_scaling: float,
mask_blur_size: float,
):
mask = np.zeros_like(image)
cascade = cv2.CascadeClassifier(classifier)
gray = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2GRAY)
faces = cascade.detectMultiScale(
gray,
minNeighbors=min_neighbors,
scaleFactor=scale_factor,
minSize=(min_size, min_size),
)
if len(faces) == 0:
return None
for x, y, w, h in faces:
top_left = np.array((x, y))
size = (np.array((w, h)) * mask_scaling).astype(int)
center = top_left + size // 2
axes = size // 2
cv2.ellipse(mask, center, axes, 0, 0, 360, (255, 255, 255), -1)
if mask_blur_size > 0:
if mask_blur_size % 2 == 0:
mask_blur_size += 1
cv2.GaussianBlur(mask, (mask_blur_size, mask_blur_size), 0, dst=mask)
return Image.fromarray(mask)
class Script(scripts.Script):
def title(self):
return "Easy Anime Face Expressions"
def show(self, is_img2img):
return is_img2img
def ui(self, _):
with gr.Row(elem_id=self.elem_id("prompt")):
expressions = gr.Textbox(
default=default_expressions,
lines=1,
label="Expression Tags",
elem_id=self.elem_id("expressions"),
)
with gr.Row(elem_id=self.elem_id("classifier")):
face_classifier = gr.Textbox(
default=default_face_classifier,
lines=1,
label="Face classifier",
elem_id=self.elem_id("face_classifier"),
)
min_neighbors = gr.Number(
default=default_min_neighbors,
label="Min neighbors",
min_value=0,
step=1,
elem_id=self.elem_id("min_neighbors"),
)
scale_factor = gr.Number(
default=default_scale_factor,
label="Scale factor",
min_value=1.0,
step=0.01,
elem_id=self.elem_id("scale_factor"),
)
min_size = gr.Number(
default=default_min_size,
label="Min size",
min_value=1,
step=1,
elem_id=self.elem_id("min_size"),
)
mask_scaling = gr.Number(
default=default_mask_scaling,
label="Mask scaling",
min_value=0.0,
step=0.1,
elem_id=self.elem_id("mask_scaling"),
)
mask_blur_size = gr.Number(
default=default_mask_blur_size,
label="Mask blur size",
min_value=0,
max_value=1,
step=1,
elem_id=self.elem_id("mask_blur_size"),
)
with gr.Row():
grid_save = gr.Checkbox(
default=opts.grid_save,
label="Save grid",
elem_id=self.elem_id("grid_save"),
)
grid_save.change(opts_setter("grid_save"), inputs=[grid_save])
save_mask = gr.Checkbox(
default=opts.save_mask,
label="Save mask",
elem_id=self.elem_id("mask_save"),
)
save_mask.change(opts_setter("save_mask"), inputs=[save_mask])
with gr.Row():
load_defaults = gr.Button(
"Load defaults",
elem_id=self.elem_id("load_defaults"),
)
load_defaults.click(
load_defaults_fn,
outputs=[
expressions,
face_classifier,
min_neighbors,
scale_factor,
min_size,
mask_scaling,
mask_blur_size,
],
)
return [
expressions,
face_classifier,
min_neighbors,
scale_factor,
min_size,
mask_scaling,
mask_blur_size,
]
def run(
self,
p: StableDiffusionProcessingImg2Img,
expressions: Any,
face_classifier: Any,
min_neighbors: Any,
scale_factor: Any,
min_size: Any,
mask_scaling: Any,
mask_blur_size: Any,
):
if face_classifier is None:
face_classifier = default_face_classifier
face_classifier_path = Path("models/face_classifier") / Path(face_classifier)
if not face_classifier_path.suffix:
face_classifier_path = face_classifier_path.with_suffix(".xml")
if not face_classifier_path.is_file():
face_classifier_path.parent.mkdir(parents=True, exist_ok=True)
print(
"Downloading lbpcascade_animeface.xml from https://github.com/nagadomi/lbpcascade_animeface"
)
url = "https://raw.githubusercontent.com/nagadomi/lbpcascade_animeface/master/lbpcascade_animeface.xml"
with open(face_classifier_path, "wb") as f:
f.write(request("GET", url).content)
fix_seed(p)
expression_list = [expression.strip() for expression in expressions.split(",")]
num_expressions = len(expression_list)
original_prompt = p.prompt
p.prompt = cast(
Any,
[
original_prompt.replace(expression_list[0], expression)
for expression in expression_list
],
)
p.image_mask = generate_face_mask(
p.init_images[0],
str(face_classifier_path),
int(min_neighbors),
scale_factor,
int(min_size),
mask_scaling,
int(mask_blur_size),
)
p.n_iter = num_expressions
p.prompt_for_display = original_prompt
processed = process_images(p)
images = [
image for image in processed.images if image.size == (p.width, p.height)
]
if len(images) != len(processed.images):
grid = image_grid(images, p.batch_size, 1)
cols, rows = num_expressions, 1
grid = draw_grid_annotations(
grid,
p.width,
p.height,
[
[GridAnnotation(text=expression)] * rows
for expression in expression_list
],
[[GridAnnotation(text="", is_active=False)] * cols],
)
processed.images = [
grid,
*images,
]
if opts.grid_save:
save_image(
grid,
p.outpath_grids,
"grid",
cast(list, p.all_seeds)[0],
cast(list, p.all_prompts)[0],
opts.grid_format,
info=processed.info,
short_filename=not opts.grid_extended_filename,
p=p,
grid=True,
)
return processed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment