Update handler.py
Browse files- handler.py +25 -35
handler.py
CHANGED
|
@@ -101,7 +101,7 @@ class EndpointHandler:
|
|
| 101 |
self.pipe.image_proj_model.to("cuda")
|
| 102 |
self.pipe.unet.to("cuda")
|
| 103 |
|
| 104 |
-
|
| 105 |
|
| 106 |
def convert_from_cv2_to_image(img: np.ndarray) -> Image:
|
| 107 |
return Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
|
|
@@ -146,41 +146,20 @@ class EndpointHandler:
|
|
| 146 |
p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
|
| 147 |
return p.replace("{prompt}", positive), n + " " + negative
|
| 148 |
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
image = Image.open(io.BytesIO(image_data))
|
| 152 |
-
return image
|
| 153 |
-
|
| 154 |
-
# Debug: Print incoming data
|
| 155 |
-
print("Incoming data:", data)
|
| 156 |
-
|
| 157 |
-
face_image_base64 = data.pop("face_image_base64", None)
|
| 158 |
-
if face_image_base64 is not None:
|
| 159 |
-
print("Using provided base64 image")
|
| 160 |
-
face_image = load_image_from_base64(face_image_base64)
|
| 161 |
-
else:
|
| 162 |
-
print("No base64 image provided, using default path or URL")
|
| 163 |
-
face_image_path = data.pop("face_image_path", "raw/input.jpg")
|
| 164 |
-
face_image = load_image(face_image_path)
|
| 165 |
-
|
| 166 |
-
# Debug: Print the used face image
|
| 167 |
-
face_image.show(title="Debug Face Image")
|
| 168 |
-
|
| 169 |
-
pose_image_path = data.pop("pose_image_path", None)
|
| 170 |
style_name = data.pop("style_name", DEFAULT_STYLE_NAME)
|
| 171 |
prompt = data.pop("inputs", "a man flying in the sky in Mars")
|
| 172 |
negative_prompt = data.pop("negative_prompt", "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green")
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
| 181 |
-
|
| 182 |
-
enhance_non_face_region = data.pop("enhance_non_face_region", False)
|
| 183 |
-
seed = data.pop("seed", 42)
|
| 184 |
|
| 185 |
self.pipe.disable_lora()
|
| 186 |
scheduler_class_name = scheduler.split("-")[0]
|
|
@@ -196,6 +175,7 @@ class EndpointHandler:
|
|
| 196 |
# apply the style template
|
| 197 |
prompt, negative_prompt = apply_style(style_name, prompt, negative_prompt)
|
| 198 |
|
|
|
|
| 199 |
face_image = resize_img(face_image, max_side=1024)
|
| 200 |
face_image_cv2 = convert_from_image_to_cv2(face_image)
|
| 201 |
height, width, _ = face_image_cv2.shape
|
|
@@ -203,6 +183,11 @@ class EndpointHandler:
|
|
| 203 |
# Extract face features
|
| 204 |
face_info = self.app.get(face_image_cv2)
|
| 205 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
face_info = sorted(
|
| 207 |
face_info,
|
| 208 |
key=lambda x: (x["bbox"][2] - x["bbox"][0]) * x["bbox"][3] - x["bbox"][1],
|
|
@@ -220,6 +205,11 @@ class EndpointHandler:
|
|
| 220 |
|
| 221 |
face_info = self.app.get(pose_image_cv2)
|
| 222 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 223 |
face_info = face_info[-1]
|
| 224 |
face_kps = draw_kps(pose_image, face_info["kps"])
|
| 225 |
|
|
@@ -247,7 +237,7 @@ class EndpointHandler:
|
|
| 247 |
for s in controlnet_selection
|
| 248 |
]
|
| 249 |
|
| 250 |
-
generator = torch.Generator(device=device).manual_seed(
|
| 251 |
|
| 252 |
print("Start inference...")
|
| 253 |
print(f"[Debug] Prompt: {prompt}, \n[Debug] Neg Prompt: {negative_prompt}")
|
|
@@ -272,4 +262,4 @@ class EndpointHandler:
|
|
| 272 |
images[0].save(buffered, format="JPEG")
|
| 273 |
img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
|
| 274 |
|
| 275 |
-
return {"generated_image_base64": img_str}
|
|
|
|
| 101 |
self.pipe.image_proj_model.to("cuda")
|
| 102 |
self.pipe.unet.to("cuda")
|
| 103 |
|
| 104 |
+
def __call__(self, data):
|
| 105 |
|
| 106 |
def convert_from_cv2_to_image(img: np.ndarray) -> Image:
|
| 107 |
return Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
|
|
|
|
| 146 |
p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
|
| 147 |
return p.replace("{prompt}", positive), n + " " + negative
|
| 148 |
|
| 149 |
+
face_image_path = data.pop("face_image_path", "https://i.ibb.co/GQzm527/examples-musk-resize.jpg")
|
| 150 |
+
pose_image_path = data.pop("pose_image_path", "https://i.ibb.co/TRCK4MS/examples-poses-pose2.jpg")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 151 |
style_name = data.pop("style_name", DEFAULT_STYLE_NAME)
|
| 152 |
prompt = data.pop("inputs", "a man flying in the sky in Mars")
|
| 153 |
negative_prompt = data.pop("negative_prompt", "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green")
|
| 154 |
+
|
| 155 |
+
identitynet_strength_ratio = 0.8
|
| 156 |
+
adapter_strength_ratio = 0.8
|
| 157 |
+
pose_strength = 0.5
|
| 158 |
+
canny_strength = 0.3
|
| 159 |
+
num_steps = 20
|
| 160 |
+
guidance_scale = 5.0
|
| 161 |
+
controlnet_selection = ["pose", "canny"]
|
| 162 |
+
scheduler = "EulerDiscreteScheduler"
|
|
|
|
|
|
|
| 163 |
|
| 164 |
self.pipe.disable_lora()
|
| 165 |
scheduler_class_name = scheduler.split("-")[0]
|
|
|
|
| 175 |
# apply the style template
|
| 176 |
prompt, negative_prompt = apply_style(style_name, prompt, negative_prompt)
|
| 177 |
|
| 178 |
+
face_image = load_image(face_image_path)
|
| 179 |
face_image = resize_img(face_image, max_side=1024)
|
| 180 |
face_image_cv2 = convert_from_image_to_cv2(face_image)
|
| 181 |
height, width, _ = face_image_cv2.shape
|
|
|
|
| 183 |
# Extract face features
|
| 184 |
face_info = self.app.get(face_image_cv2)
|
| 185 |
|
| 186 |
+
# if len(face_info) == 0:
|
| 187 |
+
# raise gr.Error(
|
| 188 |
+
# f"Unable to detect a face in the image. Please upload a different photo with a clear face."
|
| 189 |
+
# )
|
| 190 |
+
|
| 191 |
face_info = sorted(
|
| 192 |
face_info,
|
| 193 |
key=lambda x: (x["bbox"][2] - x["bbox"][0]) * x["bbox"][3] - x["bbox"][1],
|
|
|
|
| 205 |
|
| 206 |
face_info = self.app.get(pose_image_cv2)
|
| 207 |
|
| 208 |
+
# if len(face_info) == 0:
|
| 209 |
+
# raise gr.Error(
|
| 210 |
+
# f"Cannot find any face in the reference image! Please upload another person image"
|
| 211 |
+
# )
|
| 212 |
+
|
| 213 |
face_info = face_info[-1]
|
| 214 |
face_kps = draw_kps(pose_image, face_info["kps"])
|
| 215 |
|
|
|
|
| 237 |
for s in controlnet_selection
|
| 238 |
]
|
| 239 |
|
| 240 |
+
generator = torch.Generator(device=device).manual_seed(42)
|
| 241 |
|
| 242 |
print("Start inference...")
|
| 243 |
print(f"[Debug] Prompt: {prompt}, \n[Debug] Neg Prompt: {negative_prompt}")
|
|
|
|
| 262 |
images[0].save(buffered, format="JPEG")
|
| 263 |
img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
|
| 264 |
|
| 265 |
+
return {"generated_image_base64": img_str}
|