bilder-builder / app.py
Sebastiankay's picture
Update app.py
4a27b60 verified
raw
history blame
17.8 kB
import os, sys, json, re, time, base64, random, shutil
import gradio as gr
import numpy as np
import requests
from requests import Session
from requests.adapters import HTTPAdapter
from requests.packages.urllib3.util.retry import Retry
from datetime import datetime
import urllib.parse
from groq import Groq
from exif import Image
from PIL import Image as PILImage, ExifTags as PILExifTags
from io import BytesIO
import colorsys
# MARK: INIT
# MAX_SEED = np.iinfo(np.int32).max
MAX_SEED = 9999
MAX_IMAGE_SIZE = 2048
GROQ_APIKEY_PROMPTENHANCE = os.getenv("GROQ_APIKEY_PROMPTENHANCE")
API_V1 = os.getenv("API_V1")
API_V1_MODELS = urllib.parse.unquote(API_V1 + "/models")
API_V1_IMAGE = urllib.parse.unquote(API_V1 + "/imagine")
API_OLD = urllib.parse.unquote(os.getenv("API_OLD"))
# print(API_V1_MODELS)
# print(API_V1_IMAGE)
CACHE_DIR = os.path.join(os.path.dirname(__file__), "cache")
IMAGE_DIR = os.path.join(CACHE_DIR, "images")
if not os.path.exists(CACHE_DIR):
os.makedirs(CACHE_DIR)
print(f"Created cache dir on path {CACHE_DIR}")
os.makedirs(IMAGE_DIR)
print(f"Created images dir on path {IMAGE_DIR}")
RES = os.path.join(os.path.dirname(__file__), "_res")
gr.set_static_paths(paths=["_res/assets/", "_res/assets/emojis/", "_res/assets/favicons/"])
custom_css = RES + "/_custom.css"
custom_js = RES + "/_custom.js"
custom_head = f"""
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/5.9.0/css/all.min.css"/>
<script src="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/5.9.0/js/all.min.js"></script>
<link rel="apple-touch-icon" sizes="180x180" href="file=_res/assets/favicons/apple-touch-icon.png">
<link rel="icon" type="image/png" sizes="32x32" href="file=_res/assets/favicons/favicon-32x32.png">
<link rel="icon" type="image/png" sizes="16x16" href="file=_res/assets/favicons/favicon-16x16.png">
<link rel="icon" type="image/x-icon" href="file=_res/assets/favicons/favicon.ico">
<link rel="manifest" href="file=_res/assets/favicons/site.webmanifest">
"""
theme = gr.themes.Soft(
# primary_hue="orange",
radius_size="sm",
neutral_hue=gr.themes.Color(c100="#a6adc8", c200="#9399b2", c300="#7f849c", c400="#6c7086", c50="#cdd6f4", c500="#585b70", c600="#45475a", c700="#313244", c800="#1e1e2e", c900="#181825", c950="#11111b"),
)
title = "Bilder Builder"
# MARK: GET PREV. IMAGES
def get_gallery_images(dirpath):
gallery_images = [dirpath + "/" + s for s in os.listdir(dirpath) if os.path.isfile(os.path.join(dirpath, s))]
gallery_images.sort(key=lambda s: os.path.getmtime(s), reverse=True)
return gallery_images
# MARK: READ EXIF
def read_exif(image_path):
with open(image_path, "rb") as src:
img = Image(src)
img_comment = json.loads(img.user_comment)
# checking if the key exists before removing
if "concept" in img_comment:
img_comment.pop("concept")
return img_comment
def read_image_exfi_data(image_path):
print("Imagepath:", image_path)
img_exif_make, img_exif_comment = read_exif(image_path)
return None, image_path, img_exif_comment
# MARK: GROQ PROMPT ENHANCE
def groq_enhance_process(Prompt=""):
client = Groq(api_key=GROQ_APIKEY_PROMPTENHANCE)
Prompt = "random prompt" if Prompt == "" else Prompt
SYSTEMPROMPT = os.path.join(RES, "groq_systemmessage_prompt_enhance.json")
with open(SYSTEMPROMPT, "r") as f:
SYSTEMPROMPT = json.load(f)
completion = client.chat.completions.create(
model="llama-3.1-70b-versatile",
messages=[SYSTEMPROMPT, {"role": "user", "content": Prompt}],
temperature=0.8,
max_tokens=512,
top_p=0.9,
stream=False,
seed=random.randint(0, MAX_SEED),
stop=None,
)
if completion.choices[0].message.content != "":
enhanced_prompt = completion.choices[0].message.content
enhanced_prompt = re.sub(r"[\.\"]+", "", enhanced_prompt)
return enhanced_prompt
def image_get_size(image_path):
img = PILImage.open(image_path)
print("Image size:", img.size)
width, height = img.size
return width, height
# MARK: DOMINANT COLOR
def image_get_dominant_color(image_path):
img = PILImage.open(image_path)
img = img.convert("RGB")
img = img.resize((100, 100), resample=0)
pixels = list(img.getdata())
# Erzeuge eine Liste mit den Häufigkeiten der Farben
colors = []
for pixel in pixels:
r, g, b = pixel
h, s, v = colorsys.rgb_to_hsv(r / 255, g / 255, b / 255)
if v > 0.5: # Filteriere hellere Farben aus
continue
if v > 0.99: # Filteriere Weiß aus
continue
colors.append((h, s, v))
# Ermittle die dominante Farbe
dominant_color = max(colors, key=lambda x: x[2])
dominant_color_rgb = colorsys.hsv_to_rgb(dominant_color[0], dominant_color[1], dominant_color[2])
dominant_color_rgb = [int(c * 255) for c in dominant_color_rgb]
dominant_color_rgb = f"rgb({dominant_color_rgb[0]}, {dominant_color_rgb[1]}, {dominant_color_rgb[2]})"
print(dominant_color_rgb)
return dominant_color_rgb
# MARK: CLEAR COMPONENTS
def clear_components():
return None
def process(Prompt, used_model, image_width, image_height, image_ratio, image_seed, randomize_seed):
if Prompt == "":
gr.Info("Kein Prompt angegeben, es wird ein zufälliger Prompt generiert.", duration=12)
Prompt = groq_enhance_process("random prompt")
image_ratio = "9:16" if image_ratio == "" else image_ratio
used_seed = random.randint(0, MAX_SEED) if image_seed == 0 or randomize_seed else image_seed
# used_model = "turbo" if int(image_width) > 1024 or int(image_height) > 1024 else "flux" # turbo, flux
timestamp = datetime.now().strftime("%Y-%m-%d-%H-%M-%S")
filename_prompt = re.sub(r"[^\w\s-]", "", Prompt).strip().replace(" ", "_")
filename = timestamp + "_" + filename_prompt[:100] + ".png"
file_path = os.path.join(IMAGE_DIR, filename)
# Retry-Logik mit requests und Retry
session = Session()
retries = Retry(total=3, status_forcelist=[429, 500, 502, 503, 504], backoff_factor=0.3, respect_retry_after_header=True) # Max 3 Versuche # Codes, die wiederholt werden # Exponential Backoff # Retry-Header beachten
adapter = HTTPAdapter(max_retries=retries)
session.mount("https://", adapter)
session.mount("http://", adapter)
API_OLD_REQUEST_URL = f"{API_OLD}{urllib.parse.quote(Prompt)}?model=flux&width={image_width}&height={image_height}&nologo=true&enhance=false&nofeed=true&seed={used_seed}"
API_V1_REQUEST_URL = f"{API_V1_IMAGE}?prompt={urllib.parse.quote(Prompt)}&size={urllib.parse.quote(image_ratio)}&seed={used_seed}&model={used_model.lower()}"
REQUEST_URL = API_V1_REQUEST_URL if used_model.lower() != "flux-api-old" else API_OLD_REQUEST_URL
print(REQUEST_URL)
try:
response = session.get(REQUEST_URL, timeout=60)
if response.status_code == 200:
print("Imagine API Request solved")
print("Save image to: " + file_path)
img = PILImage.open(BytesIO(response.content))
img.save(file_path, "PNG")
# img_exif_comment = "" # read_exif(file_path)
img_dominant_color = image_get_dominant_color(file_path)
img_width, img_height = image_get_size(file_path)
return ({"value": Prompt, "__type__": "update"}, {"value": file_path, "__type__": "update"}, {"value": None, "visible": False, "__type__": "update"}, {"visible": True, "__type__": "update"}, {"value": Prompt, "visible": True, "__type__": "update"}, img_width, img_height, used_seed, {"value": file_path, "visible": True, "__type__": "update"}, img_dominant_color, used_seed)
else:
print("Imagine API Request ERROR")
raise gr.Error("Imagine API-Aufruf fehlgeschlagen 💥!", duration=15)
except requests.exceptions.Timeout:
raise gr.Error("⏰ Zeitüberschreitung beim API-Aufruf", duration=15)
except requests.exceptions.RequestException as e:
print(f"Unbekannter Fehler beim API-Aufruf: {e}")
raise gr.Error("Unbekannter Fehler beim API-Aufruf! 🤷‍♂️", duration=15)
# MARK: Gradio BLOCKS UI
with gr.Blocks(theme=theme, head=custom_head, css=custom_css, js=custom_js, title=title) as demo:
with gr.Row(elem_classes="row-header"):
gr.Markdown(f"""
<h1>{title}</h1>
<p><span style="font-weight: 600">LG Sebastian</span> <i class="winking-hand-emoji"></i> gib dem Space gerne ein <i class="heart-beat-emoji"></i></p>
""", elem_classes="md-header",)
with gr.Tab("Bilder Builder"):
with gr.Row():
with gr.Column(scale=2): # min_width=420,
with gr.Row():
placeholder_text = "[???] Generiert dir einen zufälligen Prompt.\n[STERN] optimiert deinen eignen Prompt.\n[RUN] generiert dein Bild."
text_prompt = gr.Textbox(label="Prompt", show_label=False, lines=12, max_lines=18, placeholder=placeholder_text, elem_id="prompt_input", elem_classes="prompt-input hide-progress", autofocus=True)
with gr.Row():
random_prompt_button = gr.Button("", variant="secondary", elem_id="random_prompt_btn", elem_classes="random-prompt-btn", icon="_res/assets/star_light_48.png")
enhance_prompt_button = gr.Button("", variant="secondary", elem_id="enhance_prompt_btn", elem_classes="enhance-prompt-btn", icon="_res/assets/star_light_48.png")
run_button = gr.Button("Erstellen", variant="primary", elem_id="run_btn", elem_classes="run-btn")
with gr.Row(elem_classes="image_size_selctor_wrapper"):
with gr.Column(scale=1):
with gr.Row():
inference_models = ["FLUX-API-OLD"] # ["FLUX", "FLUX-Realism", "FLUX-Pixel", "FLUX-Anime", "FLUX-3D", "FLUX-Disney", "ANY-DARK", "Stable-Diffusion-XL-Base", "Stable-Diffusion-XL-Lightning"]
select_model = gr.Dropdown(choices=inference_models, value="FLUX-API-OLD", label="Model", elem_id="select_model", elem_classes="select-model")
# with gr.Row():
image_width = gr.Number(label="Breite", minimum=256, maximum=MAX_IMAGE_SIZE, value=576, step=32, elem_id="image_width_selector", elem_classes="image-width-selector", scale=1, visible=False)
image_height = gr.Number(label="Höhe", minimum=256, maximum=MAX_IMAGE_SIZE, value=1024, step=32, elem_id="image_height_selector", elem_classes="image-height-selector", scale=1, visible=False)
with gr.Row():
image_ratio_buttons = gr.Radio(["9:16", "3:4", "2:3", "1:1"], value="9:16", label="Hochformat", show_label=True, info="Die gängigsten Seitenverhältnissen.", interactive=True, elem_id="image_ratio_buttons", elem_classes="image-ratio-buttons", container=True, scale=2)
switch_width_height = gr.Button("", size="sm", elem_id="switch_width_height", elem_classes="switch-ratio-btn", variant="primary", scale=1)
with gr.Column():
randomize_seed = gr.Checkbox(label="Randomize seed", value=True, elem_classes="random-seed-cb")
image_seed = gr.Slider(label="Seed", info="Jeder Seed generiert ein anderes Bild mit dem selben Prompt", minimum=0, step=1, value=42, maximum=MAX_SEED, elem_id="image_seed", elem_classes="image-seed hide-progress", interactive=False)
with gr.Column(scale=4): # min_width=600,
with gr.Row():
with gr.Column(scale=1):
with gr.Row():
output_image = gr.Image(show_label=False, min_width=320, scale=3, elem_id="output_image", elem_classes="output-image", type="filepath", format="webp")
# gallery = gr.Gallery(label="Bisher erstellte Bilder", show_label=True, value=get_gallery_images(IMAGE_DIR), elem_id="gallery", columns=[4], object_fit="cover", height="auto", interactive=False, format="webp")
with gr.Column(scale=1, visible=False, elem_classes="image-info-wrapper") as image_info_wrapper:
with gr.Group():
image_informations = gr.Markdown("""## Bildinformationen""", visible=True)
with gr.Row(elem_classes="img-size-wrapper"):
image_info_tb_width = gr.Textbox(label="Breite", lines=1, max_lines=1, interactive=False, show_copy_button=True, elem_classes="image-info-tb-width")
image_info_tb_height = gr.Textbox(label="Höhe", lines=1, max_lines=1, interactive=False, show_copy_button=True, elem_classes="image-info-tb-height")
with gr.Row(elem_classes="img-seed-wrapper"):
image_info_tb_seed = gr.Textbox(label="Seed", lines=1, max_lines=1, interactive=False, show_copy_button=True, elem_classes="image-info-tb-seed")
image_info_tb_prompt = gr.Textbox("Bild Prompt", lines=4, max_lines=8, interactive=False, elem_classes="hide-progress", show_copy_button=True, visible=False)
image_download_button = gr.DownloadButton("Bild herunterladen", value=None, elem_classes="download-button", variant="primary", visible=False)
output_url = gr.Textbox(label="Output URL", show_label=True, interactive=False, visible=False)
outpu_image_comment = gr.Json(visible=False)
output_dominant_image_color = gr.Textbox(show_label=False, elem_id="dominant_image_color", visible=True, elem_classes="output-dominant-image-color")
def switch_image_size_values(image_width, image_height):
return image_height, image_width
def switch_image_ratio_buttons(ratio_value):
ratio_value = ratio_value.split(":")
ratio_value_new = f"{int(ratio_value[1])}:{int(ratio_value[0])}"
if int(ratio_value[1]) > int(ratio_value[0]):
# Querformat
new_choises = ["16:9", "4:3", "3:2", "1:1"]
new_label = "Querformat"
elif int(ratio_value[1]) < int(ratio_value[0]):
# Hochformat
new_choises = ["9:16", "3:4", "2:3", "1:1"]
new_label = "Hochformat"
elif int(ratio_value[1]) == int(ratio_value[0]):
new_choises = image_ratio_buttons.choices
new_label = "Quadratisch"
return {"choices": new_choises, "value": ratio_value_new, "label": new_label, "__type__": "update"}
def calculate_ratio_values(image_ratio_buttons):
ratio_value = image_ratio_buttons.split(":")
if int(ratio_value[0]) > int(ratio_value[1]):
a = 1024
b = int(a * int(ratio_value[1]) / int(ratio_value[0]))
new_width = a
new_height = b
new_label = "Querformat"
elif int(ratio_value[0]) < int(ratio_value[1]):
b = 1024
a = int(b * int(ratio_value[0]) / int(ratio_value[1]))
new_width = a
new_height = b
new_label = "Hochformat"
elif int(ratio_value[0]) == int(ratio_value[1]):
new_width = 1024
new_height = 1024
new_label = "Quadratisch"
return {"label": new_label, "__type__": "update"}, new_width, new_height
switch_width_height.click(fn=switch_image_size_values, inputs=[image_width, image_height], outputs=[image_width, image_height], show_progress="hidden", show_api=False)
switch_width_height.click(fn=switch_image_ratio_buttons, inputs=[image_ratio_buttons], outputs=[image_ratio_buttons], show_progress="hidden", show_api=False)
image_ratio_buttons.input(fn=calculate_ratio_values, inputs=[image_ratio_buttons], outputs=[image_ratio_buttons, image_width, image_height], show_progress="hidden", show_api=False)
run_button.click(fn=lambda: ({"interactive": False, "__type__": "update"}, {"interactive": False, "__type__": "update"}, {"interactive": False, "__type__": "update"}), outputs=[run_button, enhance_prompt_button, random_prompt_button], show_api=False).then(fn=clear_components, outputs=[output_image], show_api=False).then(
fn=process, inputs=[text_prompt, select_model, image_width, image_height, image_ratio_buttons, image_seed, randomize_seed], outputs=[text_prompt, output_image, output_url, image_informations, image_info_tb_prompt, image_info_tb_width, image_info_tb_height, image_info_tb_seed, image_download_button, output_dominant_image_color, image_seed]
).then(fn=lambda: ({"interactive": True, "__type__": "update"}, {"interactive": True, "__type__": "update"}, {"interactive": True, "__type__": "update"}), outputs=[run_button, enhance_prompt_button, random_prompt_button], show_api=False)
randomize_seed.input(lambda x: {"interactive": False if x == True else True, "__type__": "update"}, inputs=[randomize_seed], outputs=[image_seed], show_api=False)
enhance_prompt_button.click(fn=groq_enhance_process, inputs=[text_prompt], outputs=[text_prompt], show_api=False)
random_prompt_button.click(fn=groq_enhance_process, inputs=None, outputs=[text_prompt], show_api=False)
# MARK: Gradio LAUNCH
demo.launch(show_api=True)