File size: 15,597 Bytes
360d274
 
 
 
 
 
 
9469b65
 
360d274
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44a51c4
360d274
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7d53395
 
 
 
 
fc80014
7d53395
360d274
 
 
 
 
 
 
 
 
 
 
 
 
 
44a51c4
 
 
7d53395
edb06d1
44a51c4
 
 
 
 
 
 
 
360d274
 
 
 
 
 
 
 
 
 
 
44a51c4
360d274
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44a51c4
360d274
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44a51c4
d4ecb28
 
 
 
 
 
360d274
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9469b65
 
ce49e2b
9469b65
360d274
 
 
b269831
360d274
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
from typing import Any
from diffusers import LCMScheduler
import torch
from backend.models.lcmdiffusion_setting import LCMDiffusionSetting
import numpy as np
from constants import DEVICE
from backend.models.lcmdiffusion_setting import LCMLora
# from backend.utils_sd import register_normal_pipeline, register_faster_forward, register_parallel_pipeline, seed_everything  # 1.import package
# print("我加载了")
from backend.device import is_openvino_device
from backend.openvino.pipelines import (
    get_ov_text_to_image_pipeline,
    ov_load_taesd,
    get_ov_image_to_image_pipeline,
)
from backend.pipelines.lcm import (
    get_lcm_model_pipeline,
    load_taesd,
    get_image_to_image_pipeline,
)
from backend.pipelines.lcm_lora import get_lcm_lora_pipeline
from backend.models.lcmdiffusion_setting import DiffusionTask
from image_ops import resize_pil_image
from math import ceil


class LCMTextToImage:
    def __init__(
        self,
        device: str = "cpu",
    ) -> None:
        self.pipeline = None
        self.use_openvino = False
        self.device = ""
        self.previous_model_id = None
        self.previous_use_tae_sd = False
        self.previous_use_lcm_lora = False
        self.previous_ov_model_id = ""
        self.previous_safety_checker = False
        self.previous_use_openvino = False
        self.img_to_img_pipeline = None
        self.is_openvino_init = False
        self.task_type = DiffusionTask.text_to_image
        self.torch_data_type = (
            torch.float32 if is_openvino_device() or DEVICE == "mps" else torch.float16
        )
        print(f"Torch datatype : {self.torch_data_type}")

    def _pipeline_to_device(self):
        print(f"Pipeline device : {DEVICE}")
        print(f"Pipeline dtype : {self.torch_data_type}")
        self.pipeline.to(
            torch_device=DEVICE,
            torch_dtype=self.torch_data_type,
        )

    def _add_freeu(self):
        pipeline_class = self.pipeline.__class__.__name__
        if isinstance(self.pipeline.scheduler, LCMScheduler):
            if pipeline_class == "StableDiffusionPipeline":
                print("Add FreeU - SD")
                self.pipeline.enable_freeu(
                    s1=0.9,
                    s2=0.2,
                    b1=1.2,
                    b2=1.4,
                )
            elif pipeline_class == "StableDiffusionXLPipeline":
                print("Add FreeU - SDXL")
                self.pipeline.enable_freeu(
                    s1=0.6,
                    s2=0.4,
                    b1=1.1,
                    b2=1.2,
                )

    def _update_lcm_scheduler_params(self):
        if isinstance(self.pipeline.scheduler, LCMScheduler):
            self.pipeline.scheduler = LCMScheduler.from_config(
                self.pipeline.scheduler.config,
                beta_start=0.001,
                beta_end=0.01,
            )

    def make_even(self,num):
        if num % 2 == 0:
            return num
        else:
            print("取整了")
            return num+1

    def init(
        self,
        device: str = "cpu",
        lcm_diffusion_setting: LCMDiffusionSetting = LCMDiffusionSetting(),
    ) -> None:
        self.device = device
        self.use_openvino = lcm_diffusion_setting.use_openvino
        model_id = lcm_diffusion_setting.lcm_model_id
        use_local_model = lcm_diffusion_setting.use_offline_model
        use_tiny_auto_encoder = lcm_diffusion_setting.use_tiny_auto_encoder
        use_lora = lcm_diffusion_setting.use_lcm_lora
        lcm_lora: LCMLora = lcm_diffusion_setting.lcm_lora
        ov_model_id = lcm_diffusion_setting.openvino_lcm_model_id

        if lcm_diffusion_setting.diffusion_task == DiffusionTask.image_to_image.value:
            w, h = lcm_diffusion_setting.init_image.size
            newW = lcm_diffusion_setting.image_width
            newH = self.make_even(int(h * newW / w))
            # lcm_diffusion_setting.image_height=newH
            img = lcm_diffusion_setting.init_image.resize((newW, newH)) 
            print("新图",newH,newW, lcm_diffusion_setting.image_height)
            lcm_diffusion_setting.init_image = resize_pil_image(
                img,
                lcm_diffusion_setting.image_width,
                lcm_diffusion_setting.image_height,
            )
            print("图片大小",lcm_diffusion_setting.init_image)

        if (
            self.pipeline is None
            or self.previous_model_id != model_id
            or self.previous_use_tae_sd != use_tiny_auto_encoder
            or self.previous_lcm_lora_base_id != lcm_lora.base_model_id
            or self.previous_lcm_lora_id != lcm_lora.lcm_lora_id
            or self.previous_use_lcm_lora != use_lora
            or self.previous_ov_model_id != ov_model_id
            or self.previous_safety_checker != lcm_diffusion_setting.use_safety_checker
            or self.previous_use_openvino != lcm_diffusion_setting.use_openvino
            or self.previous_task_type != lcm_diffusion_setting.diffusion_task
        ):
            if self.use_openvino and is_openvino_device():
                if self.pipeline:
                    del self.pipeline
                    self.pipeline = None
                self.is_openvino_init = True
                if (
                    lcm_diffusion_setting.diffusion_task
                    == DiffusionTask.text_to_image.value
                ):
                    print(f"***** Init Text to image (OpenVINO) - {ov_model_id} *****")
                    self.pipeline = get_ov_text_to_image_pipeline(
                        ov_model_id,
                        use_local_model,
                    )
                elif (
                    lcm_diffusion_setting.diffusion_task
                    == DiffusionTask.image_to_image.value
                ):
                    print(f"***** Image to image (OpenVINO) - {ov_model_id} *****")
                    self.pipeline = get_ov_image_to_image_pipeline(
                        ov_model_id,
                        use_local_model,
                    )
            else:
                if self.pipeline:
                    del self.pipeline
                    self.pipeline = None
                if self.img_to_img_pipeline:
                    del self.img_to_img_pipeline
                    self.img_to_img_pipeline = None

                if use_lora:
                    print(
                        f"***** Init LCM-LoRA pipeline - {lcm_lora.base_model_id} *****"
                    )
                    self.pipeline = get_lcm_lora_pipeline(
                        lcm_lora.base_model_id,
                        lcm_lora.lcm_lora_id,
                        use_local_model,
                        torch_data_type=self.torch_data_type,
                    )
                else:
                    print(f"***** Init LCM Model pipeline - {model_id} *****")
                    self.pipeline = get_lcm_model_pipeline(
                        model_id,
                        use_local_model,
                    )

                if (
                    lcm_diffusion_setting.diffusion_task
                    == DiffusionTask.image_to_image.value
                ):
                    self.img_to_img_pipeline = get_image_to_image_pipeline(
                        self.pipeline
                    )
                self._pipeline_to_device()

            if use_tiny_auto_encoder:
                if self.use_openvino and is_openvino_device():
                    print("Using Tiny Auto Encoder (OpenVINO)")
                    ov_load_taesd(
                        self.pipeline,
                        use_local_model,
                    )
                else:
                    print("Using Tiny Auto Encoder")
                    if (
                        lcm_diffusion_setting.diffusion_task
                        == DiffusionTask.text_to_image.value
                    ):
                        load_taesd(
                            self.pipeline,
                            use_local_model,
                            self.torch_data_type,
                        )
                    elif (
                        lcm_diffusion_setting.diffusion_task
                        == DiffusionTask.image_to_image.value
                    ):
                        load_taesd(
                            self.img_to_img_pipeline,
                            use_local_model,
                            self.torch_data_type,
                        )

            if (
                lcm_diffusion_setting.diffusion_task
                == DiffusionTask.image_to_image.value
                and lcm_diffusion_setting.use_openvino
            ):
                self.pipeline.scheduler = LCMScheduler.from_config(
                    self.pipeline.scheduler.config,
                )
            else:
                self._update_lcm_scheduler_params()

            if use_lora:
                self._add_freeu()

            self.previous_model_id = model_id
            self.previous_ov_model_id = ov_model_id
            self.previous_use_tae_sd = use_tiny_auto_encoder
            self.previous_lcm_lora_base_id = lcm_lora.base_model_id
            self.previous_lcm_lora_id = lcm_lora.lcm_lora_id
            self.previous_use_lcm_lora = use_lora
            self.previous_safety_checker = lcm_diffusion_setting.use_safety_checker
            self.previous_use_openvino = lcm_diffusion_setting.use_openvino
            self.previous_task_type = lcm_diffusion_setting.diffusion_task
            if (
                lcm_diffusion_setting.diffusion_task
                == DiffusionTask.text_to_image.value
            ):
                print(f"Pipeline : {self.pipeline}")
            elif (
                lcm_diffusion_setting.diffusion_task
                == DiffusionTask.image_to_image.value
            ):
                if self.use_openvino and is_openvino_device():
                    print(f"Pipeline : {self.pipeline}")
                else:
                    print(f"Pipeline : {self.img_to_img_pipeline}")

    def generate(
        self,
        lcm_diffusion_setting: LCMDiffusionSetting,
        reshape: bool = False,
    ) -> Any:
        guidance_scale = lcm_diffusion_setting.guidance_scale
        img_to_img_inference_steps = lcm_diffusion_setting.inference_steps
        check_step_value = int(
            lcm_diffusion_setting.inference_steps * lcm_diffusion_setting.strength
        )
        if (
            lcm_diffusion_setting.diffusion_task == DiffusionTask.image_to_image.value
            and check_step_value < 1
        ):
            img_to_img_inference_steps = ceil(1 / lcm_diffusion_setting.strength)
            print(
                f"Strength: {lcm_diffusion_setting.strength},{img_to_img_inference_steps}"
            )

        if lcm_diffusion_setting.use_seed:
            cur_seed = lcm_diffusion_setting.seed
            if self.use_openvino:
                np.random.seed(cur_seed)
            else:
                torch.manual_seed(cur_seed)

        is_openvino_pipe = lcm_diffusion_setting.use_openvino and is_openvino_device()
        if is_openvino_pipe:
            print("Using OpenVINO")
            if reshape and not self.is_openvino_init:
                print("Reshape and compile")
                self.pipeline.reshape(
                    batch_size=-1,
                    height=lcm_diffusion_setting.image_height,
                    width=lcm_diffusion_setting.image_width,
                    num_images_per_prompt=lcm_diffusion_setting.number_of_images,
                )
                self.pipeline.compile()

            if self.is_openvino_init:
                self.is_openvino_init = False

        if not lcm_diffusion_setting.use_safety_checker:
            self.pipeline.safety_checker = None
            if (
                lcm_diffusion_setting.diffusion_task
                == DiffusionTask.image_to_image.value
                and not is_openvino_pipe
            ):
                self.img_to_img_pipeline.safety_checker = None

        if (
            not lcm_diffusion_setting.use_lcm_lora
            and not lcm_diffusion_setting.use_openvino
            and lcm_diffusion_setting.guidance_scale != 1.0
        ):
            print("Not using LCM-LoRA so setting guidance_scale 1.0")
            guidance_scale = 1.0

        if lcm_diffusion_setting.use_openvino:
            if (
                lcm_diffusion_setting.diffusion_task
                == DiffusionTask.text_to_image.value
            ):
                result_images = self.pipeline(
                    prompt=lcm_diffusion_setting.prompt,
                    negative_prompt=lcm_diffusion_setting.negative_prompt,
                    num_inference_steps=lcm_diffusion_setting.inference_steps,
                    guidance_scale=guidance_scale,
                    width=lcm_diffusion_setting.image_width,
                    height=lcm_diffusion_setting.image_height,
                    num_images_per_prompt=lcm_diffusion_setting.number_of_images,
                ).images
            elif (
                lcm_diffusion_setting.diffusion_task
                == DiffusionTask.image_to_image.value
            ):
                # register_parallel_pipeline(self.pipeline) # 2. enable parallel. If memory is limited, replace it with  `register_normal_pipeline(pipe)`
                # register_faster_forward(self.pipeline.unet)  # 3. encoder propagation

                result_images = self.pipeline(
                    image=lcm_diffusion_setting.init_image,
                    strength=lcm_diffusion_setting.strength,
                    prompt=lcm_diffusion_setting.prompt,
                   
                    negative_prompt=lcm_diffusion_setting.negative_prompt,
                    num_inference_steps=img_to_img_inference_steps * 3,
                    guidance_scale=guidance_scale,
                    num_images_per_prompt=lcm_diffusion_setting.number_of_images,
                ).images

        else:
            if (
                lcm_diffusion_setting.diffusion_task
                == DiffusionTask.text_to_image.value
            ):
                result_images = self.pipeline(
                    prompt=lcm_diffusion_setting.prompt,
                    negative_prompt=lcm_diffusion_setting.negative_prompt,
                    num_inference_steps=lcm_diffusion_setting.inference_steps,
                    guidance_scale=guidance_scale,
                    width=lcm_diffusion_setting.image_width,
                    height=lcm_diffusion_setting.image_height,
                    num_images_per_prompt=lcm_diffusion_setting.number_of_images,
                ).images
            elif (
                lcm_diffusion_setting.diffusion_task
                == DiffusionTask.image_to_image.value
            ):
                result_images = self.img_to_img_pipeline(
                    image=lcm_diffusion_setting.init_image,
                    strength=lcm_diffusion_setting.strength,
                    prompt=lcm_diffusion_setting.prompt,
                    negative_prompt=lcm_diffusion_setting.negative_prompt,
                    num_inference_steps=img_to_img_inference_steps,
                    guidance_scale=guidance_scale,
                    width=lcm_diffusion_setting.image_width,
                    height=lcm_diffusion_setting.image_height,
                    num_images_per_prompt=lcm_diffusion_setting.number_of_images,
                ).images

        return result_images