Update app.py
Browse files
app.py
CHANGED
@@ -1,17 +1,11 @@
|
|
1 |
import os
|
2 |
-
|
3 |
-
# Install Flask if not already installed
|
4 |
-
return_code = os.system('pip install flask')
|
5 |
-
if return_code != 0:
|
6 |
-
raise RuntimeError("Failed to install Flask")
|
7 |
-
|
8 |
-
import gradio as gr
|
9 |
-
from random import randint
|
10 |
-
from all_models import models
|
11 |
from flask import Flask, request, send_file
|
12 |
from io import BytesIO
|
13 |
from PIL import Image, ImageChops
|
|
|
|
|
14 |
|
|
|
15 |
app = Flask(__name__)
|
16 |
|
17 |
# グローバルなモデル辞書
|
@@ -21,60 +15,35 @@ def load_model(model_name):
|
|
21 |
global models_load
|
22 |
if model_name not in models_load:
|
23 |
try:
|
24 |
-
|
|
|
|
|
|
|
25 |
print(f"Model {model_name} loaded successfully.")
|
26 |
-
models_load[model_name] = m
|
27 |
except Exception as error:
|
28 |
print(f"Error loading model {model_name}: {error}")
|
29 |
-
models_load[model_name] =
|
30 |
|
31 |
-
def gen_fn(model_str, prompt, negative_prompt=None, noise=None, cfg_scale=None, num_inference_steps=None
|
32 |
if model_str not in models_load:
|
33 |
load_model(model_str)
|
34 |
|
35 |
-
if model_str in models_load:
|
36 |
if noise == "random":
|
37 |
noise = str(randint(0, 99999999999))
|
38 |
full_prompt = f'{prompt} {noise}' if noise else prompt
|
39 |
|
40 |
-
|
41 |
-
print(f"Prompt: {full_prompt}, Negative Prompt: {negative_prompt}, CFG Scale: {cfg_scale}, Steps: {num_inference_steps}, Sampler: {sampler}")
|
42 |
-
|
43 |
-
# Construct the function call parameters dynamically
|
44 |
-
inputs = [full_prompt]
|
45 |
-
if negative_prompt:
|
46 |
-
inputs.append(negative_prompt)
|
47 |
-
if cfg_scale is not None:
|
48 |
-
inputs.append(cfg_scale)
|
49 |
-
if num_inference_steps is not None:
|
50 |
-
inputs.append(num_inference_steps)
|
51 |
-
if sampler:
|
52 |
-
inputs.append(sampler)
|
53 |
|
54 |
try:
|
55 |
# モデル呼び出し
|
56 |
-
result = models_load[model_str](
|
57 |
-
|
58 |
-
# Debugging result type
|
59 |
-
print(f"Result type: {type(result)}, Result: {result}")
|
60 |
|
61 |
-
# Check if result is an image or a file path
|
62 |
-
if isinstance(result, str): # Assuming result might be a file path
|
63 |
-
if os.path.exists(result):
|
64 |
-
image = Image.open(result)
|
65 |
-
else:
|
66 |
-
print(f"File path not found: {result}")
|
67 |
-
return None, 'File path not found'
|
68 |
-
elif isinstance(result, Image.Image):
|
69 |
-
image = result
|
70 |
-
else:
|
71 |
-
print("Result is not an image:", type(result))
|
72 |
-
return None, f"Unexpected result type: {type(result)}"
|
73 |
-
|
74 |
# Check if the image is completely black
|
75 |
black = Image.new('RGB', image.size, (0, 0, 0))
|
76 |
if ImageChops.difference(image, black).getbbox() is None:
|
77 |
-
return None, 'The image is completely black.
|
78 |
|
79 |
return image, None
|
80 |
|
@@ -89,21 +58,11 @@ def gen_fn(model_str, prompt, negative_prompt=None, noise=None, cfg_scale=None,
|
|
89 |
def home():
|
90 |
prompt = request.args.get('prompt', '')
|
91 |
model = request.args.get('model', '')
|
92 |
-
negative_prompt = request.args.get('Nprompt', None)
|
93 |
noise = request.args.get('noise', None)
|
94 |
-
|
95 |
-
num_inference_steps = request.args.get('steps', None)
|
96 |
-
sampler = request.args.get('sampler', None)
|
97 |
-
|
98 |
-
try:
|
99 |
-
if cfg_scale is not None:
|
100 |
-
cfg_scale = float(cfg_scale)
|
101 |
-
except ValueError:
|
102 |
-
return 'Invalid "cfg_scale" parameter. It should be a number.', 400
|
103 |
|
104 |
try:
|
105 |
-
|
106 |
-
num_inference_steps = int(num_inference_steps)
|
107 |
except ValueError:
|
108 |
return 'Invalid "steps" parameter. It should be an integer.', 400
|
109 |
|
@@ -114,12 +73,11 @@ def home():
|
|
114 |
return 'Please provide a "prompt" query parameter in the URL.', 400
|
115 |
|
116 |
# Generate the image
|
117 |
-
image, error_message = gen_fn(model, prompt,
|
118 |
if error_message:
|
119 |
return error_message, 400
|
120 |
|
121 |
-
if isinstance(image, Image.Image):
|
122 |
-
# Save image to BytesIO object
|
123 |
img_io = BytesIO()
|
124 |
image.save(img_io, format='PNG')
|
125 |
img_io.seek(0)
|
@@ -128,5 +86,4 @@ def home():
|
|
128 |
return 'Failed to generate image.', 500
|
129 |
|
130 |
if __name__ == '__main__':
|
131 |
-
|
132 |
-
app.run(host='0.0.0.0', port=7860) # Run Flask app
|
|
|
1 |
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
from flask import Flask, request, send_file
|
3 |
from io import BytesIO
|
4 |
from PIL import Image, ImageChops
|
5 |
+
from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler
|
6 |
+
import torch
|
7 |
|
8 |
+
# Flaskアプリケーションの初期化
|
9 |
app = Flask(__name__)
|
10 |
|
11 |
# グローバルなモデル辞書
|
|
|
15 |
global models_load
|
16 |
if model_name not in models_load:
|
17 |
try:
|
18 |
+
scheduler = EulerDiscreteScheduler.from_pretrained(model_name, subfolder="scheduler")
|
19 |
+
pipe = StableDiffusionPipeline.from_pretrained(model_name, scheduler=scheduler, torch_dtype=torch.float16)
|
20 |
+
pipe = pipe.to("cuda")
|
21 |
+
models_load[model_name] = pipe
|
22 |
print(f"Model {model_name} loaded successfully.")
|
|
|
23 |
except Exception as error:
|
24 |
print(f"Error loading model {model_name}: {error}")
|
25 |
+
models_load[model_name] = None
|
26 |
|
27 |
+
def gen_fn(model_str, prompt, negative_prompt=None, noise=None, cfg_scale=None, num_inference_steps=None):
|
28 |
if model_str not in models_load:
|
29 |
load_model(model_str)
|
30 |
|
31 |
+
if model_str in models_load and models_load[model_str] is not None:
|
32 |
if noise == "random":
|
33 |
noise = str(randint(0, 99999999999))
|
34 |
full_prompt = f'{prompt} {noise}' if noise else prompt
|
35 |
|
36 |
+
print(f"Prompt: {full_prompt}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
|
38 |
try:
|
39 |
# モデル呼び出し
|
40 |
+
result = models_load[model_str](full_prompt, num_inference_steps=num_inference_steps)
|
41 |
+
image = result.images[0] # 生成された画像を取得
|
|
|
|
|
42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
# Check if the image is completely black
|
44 |
black = Image.new('RGB', image.size, (0, 0, 0))
|
45 |
if ImageChops.difference(image, black).getbbox() is None:
|
46 |
+
return None, 'The image is completely black.'
|
47 |
|
48 |
return image, None
|
49 |
|
|
|
58 |
def home():
|
59 |
prompt = request.args.get('prompt', '')
|
60 |
model = request.args.get('model', '')
|
|
|
61 |
noise = request.args.get('noise', None)
|
62 |
+
num_inference_steps = request.args.get('steps', 50) # デフォルト値を設定
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
63 |
|
64 |
try:
|
65 |
+
num_inference_steps = int(num_inference_steps)
|
|
|
66 |
except ValueError:
|
67 |
return 'Invalid "steps" parameter. It should be an integer.', 400
|
68 |
|
|
|
73 |
return 'Please provide a "prompt" query parameter in the URL.', 400
|
74 |
|
75 |
# Generate the image
|
76 |
+
image, error_message = gen_fn(model, prompt, noise=noise, num_inference_steps=num_inference_steps)
|
77 |
if error_message:
|
78 |
return error_message, 400
|
79 |
|
80 |
+
if isinstance(image, Image.Image):
|
|
|
81 |
img_io = BytesIO()
|
82 |
image.save(img_io, format='PNG')
|
83 |
img_io.seek(0)
|
|
|
86 |
return 'Failed to generate image.', 500
|
87 |
|
88 |
if __name__ == '__main__':
|
89 |
+
app.run(host='0.0.0.0', port=7860)
|
|