Spaces:
Running
on
Zero
Running
on
Zero
tori29umai
commited on
Commit
•
402fe71
1
Parent(s):
3757039
app.py
Browse files
app.py
CHANGED
@@ -11,6 +11,20 @@ from utils.prompt_utils import remove_color
|
|
11 |
from utils.tagger import modelLoad, analysis
|
12 |
|
13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
def load_model(lora_dir, cn_dir):
|
15 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
16 |
dtype = torch.float16
|
@@ -29,30 +43,49 @@ def load_model(lora_dir, cn_dir):
|
|
29 |
return pipe
|
30 |
|
31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
class Img2Img:
|
33 |
def __init__(self):
|
34 |
self.setup_paths()
|
35 |
self.setup_models()
|
|
|
36 |
self.post_filter = True
|
37 |
self.tagger_model = None
|
38 |
self.input_image_path = None
|
39 |
|
40 |
-
def setup_paths(self):
|
41 |
-
self.path = os.getcwd()
|
42 |
-
self.cn_dir = f"{self.path}/controlnet"
|
43 |
-
self.tagger_dir = f"{self.path}/tagger"
|
44 |
-
self.lora_dir = f"{self.path}/lora"
|
45 |
-
os.makedirs(self.cn_dir, exist_ok=True)
|
46 |
-
os.makedirs(self.tagger_dir, exist_ok=True)
|
47 |
-
os.makedirs(self.lora_dir, exist_ok=True)
|
48 |
-
|
49 |
-
def setup_models(self):
|
50 |
-
load_cn_model(self.cn_dir)
|
51 |
-
load_cn_config(self.cn_dir)
|
52 |
-
load_tagger_model(self.tagger_dir)
|
53 |
-
load_lora_model(self.lora_dir)
|
54 |
-
|
55 |
-
|
56 |
def process_prompt_analysis(self, input_image_path):
|
57 |
if self.tagger_model is None:
|
58 |
self.tagger_model = modelLoad(self.tagger_dir)
|
@@ -63,7 +96,7 @@ class Img2Img:
|
|
63 |
return tags_list
|
64 |
|
65 |
|
66 |
-
def
|
67 |
css = """
|
68 |
#intro{
|
69 |
max-width: 32rem;
|
@@ -77,8 +110,11 @@ class Img2Img:
|
|
77 |
self.input_image_path = gr.Image(label="input_image", type='filepath')
|
78 |
self.prompt = gr.Textbox(label="prompt", lines=3)
|
79 |
self.negative_prompt = gr.Textbox(label="negative_prompt", lines=3, value="lowres, error, extra digit, fewer digits, cropped, worst quality,low quality, normal quality, jpeg artifacts, blurry")
|
|
|
80 |
prompt_analysis_button = gr.Button("prompt解析")
|
|
|
81 |
self.controlnet_scale = gr.Slider(minimum=0.5, maximum=1.25, value=1.0, step=0.01, label="線画忠実度")
|
|
|
82 |
generate_button = gr.Button("生成")
|
83 |
with gr.Column():
|
84 |
self.output_image = gr.Image(type="pil", label="出力画像")
|
@@ -96,41 +132,9 @@ class Img2Img:
|
|
96 |
inputs=[self.input_image_path, self.prompt, self.negative_prompt, self.controlnet_scale],
|
97 |
outputs=self.output_image
|
98 |
)
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
input_image_pil = Image.open(input_image_path)
|
106 |
-
base_size = input_image_pil.size
|
107 |
-
resize_image = resize_image_aspect_ratio(input_image_pil)
|
108 |
-
resize_image_size = resize_image.size
|
109 |
-
width, height = resize_image_size
|
110 |
-
white_base_pil = base_generation(resize_image.size, (255, 255, 255, 255)).convert("RGB")
|
111 |
-
generator = torch.manual_seed(0)
|
112 |
-
last_time = time.time()
|
113 |
-
|
114 |
-
output_image = pipe(
|
115 |
-
image=white_base_pil,
|
116 |
-
control_image=resize_image,
|
117 |
-
strength=1.0,
|
118 |
-
prompt=prompt,
|
119 |
-
negative_prompt=negative_prompt,
|
120 |
-
width=width,
|
121 |
-
height=height,
|
122 |
-
controlnet_conditioning_scale=float(controlnet_scale),
|
123 |
-
controlnet_start=0.0,
|
124 |
-
controlnet_end=1.0,
|
125 |
-
generator=generator,
|
126 |
-
num_inference_steps=30,
|
127 |
-
guidance_scale=8.5,
|
128 |
-
eta=1.0,
|
129 |
-
).images[0]
|
130 |
-
print(f"Time taken: {time.time() - last_time}")
|
131 |
-
output_image = output_image.resize(base_size, Image.LANCZOS)
|
132 |
-
return output_image
|
133 |
-
|
134 |
-
if __name__ == "__main__":
|
135 |
-
ui = Img2Img()
|
136 |
-
ui.launch()
|
|
|
11 |
from utils.tagger import modelLoad, analysis
|
12 |
|
13 |
|
14 |
+
|
15 |
+
path = os.getcwd()
|
16 |
+
cn_dir = f"{path}/controlnet"
|
17 |
+
tagger_dir = f"{path}/tagger"
|
18 |
+
lora_dir = f"{path}/lora"
|
19 |
+
os.makedirs(cn_dir, exist_ok=True)
|
20 |
+
os.makedirs(tagger_dir, exist_ok=True)
|
21 |
+
os.makedirs(lora_dir, exist_ok=True)
|
22 |
+
|
23 |
+
load_cn_model(cn_dir)
|
24 |
+
load_cn_config(cn_dir)
|
25 |
+
load_tagger_model(tagger_dir)
|
26 |
+
load_lora_model(lora_dir)
|
27 |
+
|
28 |
def load_model(lora_dir, cn_dir):
|
29 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
30 |
dtype = torch.float16
|
|
|
43 |
return pipe
|
44 |
|
45 |
|
46 |
+
@spaces.GPU
|
47 |
+
def predict(input_image_path, prompt, negative_prompt, controlnet_scale):
|
48 |
+
pipe = load_model(lora_dir, cn_dir)
|
49 |
+
input_image_pil = Image.open(input_image_path)
|
50 |
+
base_size = input_image_pil.size
|
51 |
+
resize_image = resize_image_aspect_ratio(input_image_pil)
|
52 |
+
resize_image_size = resize_image.size
|
53 |
+
width, height = resize_image_size
|
54 |
+
white_base_pil = base_generation(resize_image.size, (255, 255, 255, 255)).convert("RGB")
|
55 |
+
generator = torch.manual_seed(0)
|
56 |
+
last_time = time.time()
|
57 |
+
|
58 |
+
output_image = pipe(
|
59 |
+
image=white_base_pil,
|
60 |
+
control_image=resize_image,
|
61 |
+
strength=1.0,
|
62 |
+
prompt=prompt,
|
63 |
+
negative_prompt = negative_prompt,
|
64 |
+
width=width,
|
65 |
+
height=height,
|
66 |
+
controlnet_conditioning_scale=float(controlnet_scale),
|
67 |
+
controlnet_start=0.0,
|
68 |
+
controlnet_end=1.0,
|
69 |
+
generator=generator,
|
70 |
+
num_inference_steps=30,
|
71 |
+
guidance_scale=8.5,
|
72 |
+
eta=1.0,
|
73 |
+
).images[0]
|
74 |
+
print(f"Time taken: {time.time() - last_time}")
|
75 |
+
output_image = output_image.resize(base_size, Image.LANCZOS)
|
76 |
+
return output_image
|
77 |
+
|
78 |
+
|
79 |
+
|
80 |
class Img2Img:
|
81 |
def __init__(self):
|
82 |
self.setup_paths()
|
83 |
self.setup_models()
|
84 |
+
self.demo = self.layout()
|
85 |
self.post_filter = True
|
86 |
self.tagger_model = None
|
87 |
self.input_image_path = None
|
88 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
def process_prompt_analysis(self, input_image_path):
|
90 |
if self.tagger_model is None:
|
91 |
self.tagger_model = modelLoad(self.tagger_dir)
|
|
|
96 |
return tags_list
|
97 |
|
98 |
|
99 |
+
def layout(self):
|
100 |
css = """
|
101 |
#intro{
|
102 |
max-width: 32rem;
|
|
|
110 |
self.input_image_path = gr.Image(label="input_image", type='filepath')
|
111 |
self.prompt = gr.Textbox(label="prompt", lines=3)
|
112 |
self.negative_prompt = gr.Textbox(label="negative_prompt", lines=3, value="lowres, error, extra digit, fewer digits, cropped, worst quality,low quality, normal quality, jpeg artifacts, blurry")
|
113 |
+
|
114 |
prompt_analysis_button = gr.Button("prompt解析")
|
115 |
+
|
116 |
self.controlnet_scale = gr.Slider(minimum=0.5, maximum=1.25, value=1.0, step=0.01, label="線画忠実度")
|
117 |
+
|
118 |
generate_button = gr.Button("生成")
|
119 |
with gr.Column():
|
120 |
self.output_image = gr.Image(type="pil", label="出力画像")
|
|
|
132 |
inputs=[self.input_image_path, self.prompt, self.negative_prompt, self.controlnet_scale],
|
133 |
outputs=self.output_image
|
134 |
)
|
135 |
+
return demo
|
136 |
+
|
137 |
+
|
138 |
+
|
139 |
+
img2img = Img2Img()
|
140 |
+
img2img.demo.launch(share=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|