Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -1,54 +1,69 @@
|
|
1 |
import gradio as gr
|
2 |
import numpy as np
|
3 |
import random
|
4 |
-
import spaces
|
5 |
from diffusers import DiffusionPipeline, DPMSolverSDEScheduler
|
6 |
import torch
|
7 |
from huggingface_hub import hf_hub_download
|
8 |
from ultralytics import YOLO
|
9 |
-
import cv2
|
10 |
from PIL import Image
|
|
|
11 |
|
12 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
13 |
-
model_repo_id = "John6666/wai-ani-nsfw-ponyxl-v8-sdxl"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
-
|
16 |
-
pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch.float16)
|
17 |
pipe.scheduler = DPMSolverSDEScheduler.from_config(pipe.scheduler.config, algorithm_type="dpmsolver++", solver_order=2, use_karras_sigmas=True)
|
18 |
pipe = pipe.to(device)
|
19 |
|
20 |
MAX_SEED = np.iinfo(np.int32).max
|
21 |
MAX_IMAGE_SIZE = 1024
|
22 |
|
23 |
-
|
24 |
-
|
25 |
-
yolo_model = YOLO(yolo_model_path)
|
26 |
-
|
27 |
-
def fix_eyes_with_adetailer(image):
|
28 |
-
# Convert PIL image to OpenCV format for YOLO
|
29 |
img = np.array(image)
|
30 |
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
|
31 |
|
32 |
-
#
|
33 |
results = yolo_model(img)
|
34 |
|
35 |
-
|
36 |
-
|
37 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
|
39 |
-
# Convert
|
40 |
-
|
41 |
-
return
|
42 |
|
43 |
-
@spaces.GPU
|
44 |
def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, progress=gr.Progress(track_tqdm=True)):
|
45 |
-
|
46 |
if randomize_seed:
|
47 |
seed = random.randint(0, MAX_SEED)
|
48 |
|
49 |
generator = torch.Generator().manual_seed(seed)
|
50 |
|
51 |
-
# Generate the initial image with the diffusion model
|
52 |
image = pipe(
|
53 |
prompt=prompt,
|
54 |
negative_prompt=negative_prompt,
|
@@ -58,9 +73,9 @@ def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance
|
|
58 |
height=height,
|
59 |
generator=generator
|
60 |
).images[0]
|
61 |
-
|
62 |
-
#
|
63 |
-
corrected_image =
|
64 |
|
65 |
return corrected_image, seed
|
66 |
|
@@ -70,94 +85,41 @@ examples = [
|
|
70 |
"A delicious ceviche cheesecake slice",
|
71 |
]
|
72 |
|
73 |
-
css=
|
|
|
|
|
|
|
|
|
|
|
74 |
|
75 |
with gr.Blocks(css=css) as demo:
|
76 |
-
|
77 |
with gr.Column(elem_id="col-container"):
|
78 |
-
gr.Markdown(
|
79 |
-
# Text-to-Image Gradio Template
|
80 |
-
""")
|
81 |
|
82 |
with gr.Row():
|
83 |
-
|
84 |
-
prompt = gr.Text(
|
85 |
-
label="Prompt",
|
86 |
-
show_label=False,
|
87 |
-
max_lines=1,
|
88 |
-
placeholder="Enter your prompt",
|
89 |
-
container=False,
|
90 |
-
)
|
91 |
-
|
92 |
run_button = gr.Button("Run", scale=0)
|
93 |
|
94 |
result = gr.Image(label="Result", show_label=False)
|
95 |
|
96 |
with gr.Accordion("Advanced Settings", open=False):
|
97 |
-
|
98 |
-
|
99 |
-
label="Negative prompt",
|
100 |
-
max_lines=1,
|
101 |
-
placeholder="Enter a negative prompt",
|
102 |
-
visible=False,
|
103 |
-
)
|
104 |
-
|
105 |
-
seed = gr.Slider(
|
106 |
-
label="Seed",
|
107 |
-
minimum=0,
|
108 |
-
maximum=MAX_SEED,
|
109 |
-
step=1,
|
110 |
-
value=0,
|
111 |
-
)
|
112 |
-
|
113 |
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
114 |
|
115 |
with gr.Row():
|
|
|
|
|
116 |
|
117 |
-
width = gr.Slider(
|
118 |
-
label="Width",
|
119 |
-
minimum=256,
|
120 |
-
maximum=MAX_IMAGE_SIZE,
|
121 |
-
step=32,
|
122 |
-
value=1024, #Replace with defaults that work for your model
|
123 |
-
)
|
124 |
-
|
125 |
-
height = gr.Slider(
|
126 |
-
label="Height",
|
127 |
-
minimum=256,
|
128 |
-
maximum=MAX_IMAGE_SIZE,
|
129 |
-
step=32,
|
130 |
-
value=1024, #Replace with defaults that work for your model
|
131 |
-
)
|
132 |
-
|
133 |
with gr.Row():
|
134 |
-
|
135 |
-
|
136 |
-
label="Guidance scale",
|
137 |
-
minimum=0.0,
|
138 |
-
maximum=10.0,
|
139 |
-
step=0.1,
|
140 |
-
value=0.0, #Replace with defaults that work for your model
|
141 |
-
)
|
142 |
-
|
143 |
-
num_inference_steps = gr.Slider(
|
144 |
-
label="Number of inference steps",
|
145 |
-
minimum=1,
|
146 |
-
maximum=50,
|
147 |
-
step=1,
|
148 |
-
value=2, #Replace with defaults that work for your model
|
149 |
-
)
|
150 |
|
151 |
-
gr.Examples(
|
152 |
-
examples=examples,
|
153 |
-
inputs=[prompt]
|
154 |
-
)
|
155 |
|
156 |
-
gr.on(
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
outputs=[result, seed]
|
161 |
-
)
|
162 |
|
163 |
demo.queue().launch()
|
|
|
1 |
import gradio as gr
|
2 |
import numpy as np
|
3 |
import random
|
4 |
+
import spaces
|
5 |
from diffusers import DiffusionPipeline, DPMSolverSDEScheduler
|
6 |
import torch
|
7 |
from huggingface_hub import hf_hub_download
|
8 |
from ultralytics import YOLO
|
|
|
9 |
from PIL import Image
|
10 |
+
import cv2
|
11 |
|
12 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
13 |
+
model_repo_id = "John6666/wai-ani-nsfw-ponyxl-v8-sdxl"
|
14 |
+
adetailer_model_id = "Bingsu/adetailer" # Your ADetailer model
|
15 |
+
|
16 |
+
# Load the YOLO model for face detection
|
17 |
+
yolo_model_path = hf_hub_download(adetailer_model_id, "face_yolov8n.pt")
|
18 |
+
yolo_model = YOLO(yolo_model_path)
|
19 |
+
|
20 |
+
if torch.cuda.is_available():
|
21 |
+
torch_dtype = torch.float16
|
22 |
+
else:
|
23 |
+
torch_dtype = torch.float32
|
24 |
|
25 |
+
pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
|
|
|
26 |
pipe.scheduler = DPMSolverSDEScheduler.from_config(pipe.scheduler.config, algorithm_type="dpmsolver++", solver_order=2, use_karras_sigmas=True)
|
27 |
pipe = pipe.to(device)
|
28 |
|
29 |
MAX_SEED = np.iinfo(np.int32).max
|
30 |
MAX_IMAGE_SIZE = 1024
|
31 |
|
32 |
+
def correct_anime_face(image):
|
33 |
+
# Convert to OpenCV format
|
|
|
|
|
|
|
|
|
34 |
img = np.array(image)
|
35 |
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
|
36 |
|
37 |
+
# Detect faces
|
38 |
results = yolo_model(img)
|
39 |
|
40 |
+
for detection in results[0].boxes:
|
41 |
+
x1, y1, x2, y2 = map(int, detection.xyxy[0].tolist())
|
42 |
+
|
43 |
+
# Crop the face region
|
44 |
+
face = img[y1:y2, x1:x2]
|
45 |
+
face_pil = Image.fromarray(cv2.cvtColor(face, cv2.COLOR_BGR2RGB))
|
46 |
+
|
47 |
+
# Prompt for the correction model
|
48 |
+
prompt = "Enhance this anime character's face, fix eyes and make features more vivid."
|
49 |
+
|
50 |
+
# Process the face with the anime correction model
|
51 |
+
corrected_face = pipe(prompt=prompt, image=face_pil).images[0] # Replace with your correction model
|
52 |
+
|
53 |
+
# Place the corrected face back into the original image
|
54 |
+
img[y1:y2, x1:x2] = np.array(corrected_face)
|
55 |
|
56 |
+
# Convert back to PIL
|
57 |
+
final_image = Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
|
58 |
+
return final_image
|
59 |
|
60 |
+
@spaces.GPU
|
61 |
def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, progress=gr.Progress(track_tqdm=True)):
|
|
|
62 |
if randomize_seed:
|
63 |
seed = random.randint(0, MAX_SEED)
|
64 |
|
65 |
generator = torch.Generator().manual_seed(seed)
|
66 |
|
|
|
67 |
image = pipe(
|
68 |
prompt=prompt,
|
69 |
negative_prompt=negative_prompt,
|
|
|
73 |
height=height,
|
74 |
generator=generator
|
75 |
).images[0]
|
76 |
+
|
77 |
+
# Correct anime face in the generated image
|
78 |
+
corrected_image = correct_anime_face(image)
|
79 |
|
80 |
return corrected_image, seed
|
81 |
|
|
|
85 |
"A delicious ceviche cheesecake slice",
|
86 |
]
|
87 |
|
88 |
+
css = """
|
89 |
+
#col-container {
|
90 |
+
margin: 0 auto;
|
91 |
+
max-width: 640px;
|
92 |
+
}
|
93 |
+
"""
|
94 |
|
95 |
with gr.Blocks(css=css) as demo:
|
|
|
96 |
with gr.Column(elem_id="col-container"):
|
97 |
+
gr.Markdown("# Text-to-Image Gradio Template")
|
|
|
|
|
98 |
|
99 |
with gr.Row():
|
100 |
+
prompt = gr.Text(label="Prompt", show_label=False, max_lines=1, placeholder="Enter your prompt", container=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
run_button = gr.Button("Run", scale=0)
|
102 |
|
103 |
result = gr.Image(label="Result", show_label=False)
|
104 |
|
105 |
with gr.Accordion("Advanced Settings", open=False):
|
106 |
+
negative_prompt = gr.Text(label="Negative prompt", max_lines=1, placeholder="Enter a negative prompt", visible=False)
|
107 |
+
seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
108 |
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
109 |
|
110 |
with gr.Row():
|
111 |
+
width = gr.Slider(label="Width", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024)
|
112 |
+
height = gr.Slider(label="Height", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024)
|
113 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
114 |
with gr.Row():
|
115 |
+
guidance_scale = gr.Slider(label="Guidance scale", minimum=0.0, maximum=10.0, step=0.1, value=0.0)
|
116 |
+
num_inference_steps = gr.Slider(label="Number of inference steps", minimum=1, maximum=50, step=1, value=2)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
|
118 |
+
gr.Examples(examples=examples, inputs=[prompt])
|
|
|
|
|
|
|
119 |
|
120 |
+
gr.on(triggers=[run_button.click, prompt.submit],
|
121 |
+
fn=infer,
|
122 |
+
inputs=[prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
|
123 |
+
outputs=[result, seed])
|
|
|
|
|
124 |
|
125 |
demo.queue().launch()
|