File size: 10,776 Bytes
86248f3
19b3da3
 
 
 
86248f3
a3d6c18
 
 
 
 
 
19b3da3
86248f3
19b3da3
 
 
 
a3d6c18
 
 
19b3da3
 
 
 
 
 
 
 
 
a3d6c18
19b3da3
a3d6c18
86248f3
19b3da3
 
 
 
a3d6c18
19b3da3
 
 
 
a3d6c18
19b3da3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86248f3
19b3da3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86248f3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19b3da3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86248f3
 
 
a3d6c18
86248f3
 
 
 
 
 
 
 
 
 
 
 
 
a3d6c18
 
 
 
 
86248f3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a3d6c18
86248f3
 
 
 
 
 
 
 
 
 
 
 
19b3da3
 
 
 
 
 
86248f3
 
 
 
 
a3d6c18
 
86248f3
 
 
 
19b3da3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
from typing import List, Union

import cv2
import numpy as np
import torch
from controlnet_aux import HEDdetector, LineartDetector, OpenposeDetector
from diffusers import (
    ControlNetModel,
    DiffusionPipeline,
    StableDiffusionControlNetPipeline,
    UniPCMultistepScheduler,
)
from PIL import Image
from torch.nn import Linear
from tqdm import gui

from internals.data.result import Result
from internals.pipelines.commons import AbstractPipeline
from internals.pipelines.tileUpscalePipeline import (
    StableDiffusionControlNetImg2ImgPipeline,
)
from internals.util.cache import clear_cuda_and_gc
from internals.util.commons import download_image


class ControlNet(AbstractPipeline):
    __current_task_name = ""

    def load(self, model_dir: str):
        # we will load canny by default
        self.load_scribble()

        # controlnet pipeline for tile upscaler
        pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
            model_dir,
            controlnet=self.controlnet,
            torch_dtype=torch.float16,
        ).to("cuda")
        # pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
        pipe.enable_model_cpu_offload()
        pipe.enable_xformers_memory_efficient_attention()
        self.pipe = pipe

        # controlnet pipeline for canny and pose
        pipe2 = StableDiffusionControlNetPipeline(**pipe.components).to("cuda")
        pipe2.scheduler = UniPCMultistepScheduler.from_config(pipe2.scheduler.config)
        pipe2.enable_xformers_memory_efficient_attention()
        self.pipe2 = pipe2

    def load_canny(self):
        if self.__current_task_name == "canny":
            return
        canny = ControlNetModel.from_pretrained(
            "lllyasviel/control_v11p_sd15_canny", torch_dtype=torch.float16
        ).to("cuda")
        self.__current_task_name = "canny"
        self.controlnet = canny
        if hasattr(self, "pipe"):
            self.pipe.controlnet = canny
        if hasattr(self, "pipe2"):
            self.pipe2.controlnet = canny
        clear_cuda_and_gc()

    def load_pose(self):
        if self.__current_task_name == "pose":
            return
        pose = ControlNetModel.from_pretrained(
            "lllyasviel/sd-controlnet-openpose", torch_dtype=torch.float16
        ).to("cuda")
        self.__current_task_name = "pose"
        self.controlnet = pose
        if hasattr(self, "pipe"):
            self.pipe.controlnet = pose
        if hasattr(self, "pipe2"):
            self.pipe2.controlnet = pose
        clear_cuda_and_gc()

    def load_tile_upscaler(self):
        if self.__current_task_name == "tile_upscaler":
            return
        tile_upscaler = ControlNetModel.from_pretrained(
            "lllyasviel/control_v11f1e_sd15_tile", torch_dtype=torch.float16
        ).to("cuda")
        self.__current_task_name = "tile_upscaler"
        self.controlnet = tile_upscaler
        if hasattr(self, "pipe"):
            self.pipe.controlnet = tile_upscaler
        if hasattr(self, "pipe2"):
            self.pipe2.controlnet = tile_upscaler
        clear_cuda_and_gc()

    def load_scribble(self):
        if self.__current_task_name == "scribble":
            return
        scribble = ControlNetModel.from_pretrained(
            "lllyasviel/control_v11p_sd15_scribble", torch_dtype=torch.float16
        ).to("cuda")
        self.__current_task_name = "scribble"
        self.controlnet = scribble
        if hasattr(self, "pipe"):
            self.pipe.controlnet = scribble
        if hasattr(self, "pipe2"):
            self.pipe2.controlnet = scribble
        clear_cuda_and_gc()

    def load_linearart(self):
        if self.__current_task_name == "linearart":
            return
        linearart = ControlNetModel.from_pretrained(
            "ControlNet-1-1-preview/control_v11p_sd15_lineart",
            torch_dtype=torch.float16,
        ).to("cuda")
        self.__current_task_name = "linearart"
        self.controlnet = linearart
        if hasattr(self, "pipe"):
            self.pipe.controlnet = linearart
        if hasattr(self, "pipe2"):
            self.pipe2.controlnet = linearart
        clear_cuda_and_gc()

    def cleanup(self):
        self.pipe.controlnet = None
        self.pipe2.controlnet = None
        self.controlnet = None
        self.__current_task_name = ""

        clear_cuda_and_gc()

    @torch.inference_mode()
    def process_canny(
        self,
        prompt: List[str],
        imageUrl: str,
        seed: int,
        steps: int,
        negative_prompt: List[str],
        guidance_scale: float,
        height: int,
        width: int,
    ):
        if self.__current_task_name != "canny":
            raise Exception("ControlNet is not loaded with canny model")

        torch.manual_seed(seed)

        init_image = download_image(imageUrl).resize((width, height))
        init_image = self.__canny_detect_edge(init_image)

        result = self.pipe2.__call__(
            prompt=prompt,
            image=init_image,
            guidance_scale=guidance_scale,
            num_images_per_prompt=1,
            negative_prompt=negative_prompt,
            num_inference_steps=steps,
            height=height,
            width=width,
        )
        return Result.from_result(result)

    @torch.inference_mode()
    def process_pose(
        self,
        prompt: List[str],
        image: List[Image.Image],
        seed: int,
        steps: int,
        guidance_scale: float,
        negative_prompt: List[str],
        height: int,
        width: int,
    ):
        if self.__current_task_name != "pose":
            raise Exception("ControlNet is not loaded with pose model")

        torch.manual_seed(seed)

        result = self.pipe2.__call__(
            prompt=prompt,
            image=image,
            num_images_per_prompt=1,
            num_inference_steps=steps,
            negative_prompt=negative_prompt,
            guidance_scale=guidance_scale,
            height=height,
            width=width,
        )
        return Result.from_result(result)

    @torch.inference_mode()
    def process_tile_upscaler(
        self,
        imageUrl: str,
        prompt: str,
        negative_prompt: str,
        steps: int,
        seed: int,
        height: int,
        width: int,
        resize_dimension: int,
        guidance_scale: float,
    ):
        if self.__current_task_name != "tile_upscaler":
            raise Exception("ControlNet is not loaded with tile_upscaler model")

        torch.manual_seed(seed)

        init_image = download_image(imageUrl).resize((width, height))
        condition_image = self.__resize_for_condition_image(
            init_image, resize_dimension
        )

        result = self.pipe.__call__(
            image=condition_image,
            prompt=prompt,
            controlnet_conditioning_image=condition_image,
            num_inference_steps=steps,
            negative_prompt=negative_prompt,
            height=condition_image.size[1],
            width=condition_image.size[0],
            guidance_scale=guidance_scale,
        )
        return Result.from_result(result)

    @torch.inference_mode()
    def process_scribble(
        self,
        imageUrl: Union[str, Image.Image],
        prompt: Union[str, List[str]],
        negative_prompt: Union[str, List[str]],
        steps: int,
        seed: int,
        height: int,
        width: int,
        guidance_scale: float = 7.5,
    ):
        if self.__current_task_name != "scribble":
            raise Exception("ControlNet is not loaded with scribble model")

        torch.manual_seed(seed)

        if isinstance(imageUrl, Image.Image):
            init_image = imageUrl.resize((width, height))
        else:
            init_image = download_image(imageUrl).resize((width, height))

        condition_image = self.__scribble_condition_image(init_image)

        result = self.pipe2.__call__(
            image=condition_image,
            prompt=prompt,
            num_inference_steps=steps,
            negative_prompt=negative_prompt,
            height=height,
            width=width,
            guidance_scale=guidance_scale,
        )
        return Result.from_result(result)

    @torch.inference_mode()
    def process_linearart(
        self,
        imageUrl: str,
        prompt: Union[str, List[str]],
        negative_prompt: Union[str, List[str]],
        steps: int,
        seed: int,
        height: int,
        width: int,
        guidance_scale: float = 7.5,
    ):
        if self.__current_task_name != "linearart":
            raise Exception("ControlNet is not loaded with linearart model")

        torch.manual_seed(seed)

        init_image = download_image(imageUrl).resize((width, height))
        condition_image = ControlNet.linearart_condition_image(init_image)

        result = self.pipe2.__call__(
            image=condition_image,
            prompt=prompt,
            num_inference_steps=steps,
            negative_prompt=negative_prompt,
            height=height,
            width=width,
            guidance_scale=guidance_scale,
        )
        return Result.from_result(result)

    def detect_pose(self, imageUrl: str) -> Image.Image:
        detector = OpenposeDetector.from_pretrained("lllyasviel/ControlNet")
        image = download_image(imageUrl)
        image = detector.__call__(image, hand_and_face=True)
        return image

    def __scribble_condition_image(self, image: Image.Image) -> Image.Image:
        processor = HEDdetector.from_pretrained("lllyasviel/Annotators")
        image = processor.__call__(input_image=image, scribble=True)
        return image

    @staticmethod
    def linearart_condition_image(image: Image.Image) -> Image.Image:
        processor = LineartDetector.from_pretrained("lllyasviel/Annotators")
        image = processor.__call__(input_image=image)
        return image

    def __canny_detect_edge(self, image: Image.Image) -> Image.Image:
        image_array = np.array(image)

        low_threshold = 100
        high_threshold = 200

        image_array = cv2.Canny(image_array, low_threshold, high_threshold)
        image_array = image_array[:, :, None]
        image_array = np.concatenate([image_array, image_array, image_array], axis=2)
        canny_image = Image.fromarray(image_array)
        return canny_image

    def __resize_for_condition_image(self, image: Image.Image, resolution: int):
        input_image = image.convert("RGB")
        W, H = input_image.size
        k = float(resolution) / min(W, H)
        H *= k
        W *= k
        H = int(round(H / 64.0)) * 64
        W = int(round(W / 64.0)) * 64
        img = input_image.resize((W, H), resample=Image.LANCZOS)
        return img