File size: 29,750 Bytes
ae97a7d
e1c7172
0223854
ae97a7d
0223854
 
e1c7172
0223854
 
 
 
ae97a7d
0223854
 
 
ae97a7d
0223854
 
 
 
 
 
 
ae97a7d
0223854
 
ae97a7d
 
 
 
 
 
0223854
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ae97a7d
 
0223854
 
ae97a7d
 
 
0223854
ae97a7d
0223854
 
 
 
ae97a7d
 
 
 
 
 
24081dd
ae97a7d
 
 
 
 
0223854
 
 
 
 
ae97a7d
 
 
 
 
 
 
 
 
0223854
ae97a7d
 
 
 
 
 
 
 
 
 
0223854
 
ae97a7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0223854
 
 
 
ae97a7d
24081dd
ae97a7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e1c7172
 
ae97a7d
0223854
 
 
 
ae97a7d
0223854
 
ae97a7d
e1c7172
 
ae97a7d
 
9e97363
 
e1c7172
 
ae97a7d
 
0223854
 
 
 
ae97a7d
 
0223854
 
ae97a7d
0223854
 
ae97a7d
0223854
 
ae97a7d
0223854
 
ae97a7d
 
 
0223854
 
ae97a7d
 
0223854
ae97a7d
 
 
 
 
 
0223854
 
 
 
 
 
 
 
 
ae97a7d
0223854
 
 
 
ae97a7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0223854
ae97a7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e1c7172
ae97a7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0223854
ae97a7d
 
 
 
 
 
 
 
0223854
ae97a7d
0223854
ae97a7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0223854
 
 
 
 
 
24081dd
c4e73e7
ae97a7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631


import os
import json
import torch
import random
import copy

import gradio as gr
from glob import glob
from omegaconf import OmegaConf
from datetime import datetime
from safetensors import safe_open

from diffusers import AutoencoderKL
from diffusers import DDIMScheduler, EulerDiscreteScheduler, PNDMScheduler
from diffusers.utils.import_utils import is_xformers_available
from transformers import CLIPTextModel, CLIPTokenizer

from animatediff.models.unet import UNet3DConditionModel
from animatediff.pipelines.pipeline_animation import AnimationPipeline
from animatediff.utils.util import save_videos_grid
from animatediff.utils.convert_from_ckpt import convert_ldm_unet_checkpoint, convert_ldm_clip_checkpoint, convert_ldm_vae_checkpoint
from animatediff.utils.convert_lora_safetensor_to_diffusers import convert_lora


sample_idx     = 0
scheduler_dict = {
    "Euler": EulerDiscreteScheduler,
    "PNDM": PNDMScheduler,
    "DDIM": DDIMScheduler,
}

css = """
.toolbutton {
    margin-buttom: 0em 0em 0em 0em;
    max-width: 2.5em;
    min-width: 2.5em !important;
    height: 2.5em;
}
"""

class AnimateController:
    def __init__(self):
        
        # config dirs
        self.basedir                = os.getcwd()
        self.stable_diffusion_dir   = os.path.join(self.basedir, "models", "StableDiffusion")
        self.motion_module_dir      = os.path.join(self.basedir, "models", "Motion_Module")
        self.personalized_model_dir = os.path.join(self.basedir, "models", "DreamBooth_LoRA")
        self.savedir                = os.path.join(self.basedir, "samples", datetime.now().strftime("Gradio-%Y-%m-%dT%H-%M-%S"))
        self.savedir_sample         = os.path.join(self.savedir, "sample")
        os.makedirs(self.savedir, exist_ok=True)

        self.stable_diffusion_list   = []
        self.motion_module_list      = []
        self.personalized_model_list = []
        
        self.refresh_stable_diffusion()
        self.refresh_motion_module()
        self.refresh_personalized_model()
        
        # config models
        self.tokenizer             = None
        self.text_encoder          = None
        self.vae                   = None
        self.unet                  = None
        self.pipeline              = None
        self.lora_model_state_dict = {}
        
        self.inference_config      = OmegaConf.load("configs/inference/inference.yaml")

    def refresh_stable_diffusion(self):
        self.stable_diffusion_list = glob(os.path.join(self.stable_diffusion_dir, "*/"))

    def refresh_motion_module(self):
        motion_module_list = glob(os.path.join(self.motion_module_dir, "*.ckpt"))
        self.motion_module_list = [os.path.basename(p) for p in motion_module_list]

    def refresh_personalized_model(self):
        personalized_model_list = glob(os.path.join(self.personalized_model_dir, "*.safetensors"))
        self.personalized_model_list = [os.path.basename(p) for p in personalized_model_list]

    def update_stable_diffusion(self, stable_diffusion_dropdown):
        self.tokenizer = CLIPTokenizer.from_pretrained(stable_diffusion_dropdown, subfolder="tokenizer")
        self.text_encoder = CLIPTextModel.from_pretrained(stable_diffusion_dropdown, subfolder="text_encoder").cuda()
        self.vae = AutoencoderKL.from_pretrained(stable_diffusion_dropdown, subfolder="vae").cuda()
        self.unet = UNet3DConditionModel.from_pretrained_2d(stable_diffusion_dropdown, subfolder="unet", unet_additional_kwargs=OmegaConf.to_container(self.inference_config.unet_additional_kwargs)).cuda()
        return gr.Dropdown.update()

    def update_motion_module(self, motion_module_dropdown):
        if self.unet is None:
            gr.Info(f"Please select a pretrained model path.")
            return gr.Dropdown.update(value=None)
        else:
            motion_module_dropdown = os.path.join(self.motion_module_dir, motion_module_dropdown)
            motion_module_state_dict = torch.load(motion_module_dropdown, map_location="cpu")
            missing, unexpected = self.unet.load_state_dict(motion_module_state_dict, strict=False)
            assert len(unexpected) == 0
            return gr.Dropdown.update()

    def update_base_model(self, base_model_dropdown):
        if self.unet is None:
            gr.Info(f"Please select a pretrained model path.")
            return gr.Dropdown.update(value=None)
        else:
            base_model_dropdown = os.path.join(self.personalized_model_dir, base_model_dropdown)
            base_model_state_dict = {}
            with safe_open(base_model_dropdown, framework="pt", device="cpu") as f:
                for key in f.keys():
                    base_model_state_dict[key] = f.get_tensor(key)
                    
            converted_vae_checkpoint = convert_ldm_vae_checkpoint(base_model_state_dict, self.vae.config)
            self.vae.load_state_dict(converted_vae_checkpoint)

            converted_unet_checkpoint = convert_ldm_unet_checkpoint(base_model_state_dict, self.unet.config)
            self.unet.load_state_dict(converted_unet_checkpoint, strict=False)

            self.text_encoder = convert_ldm_clip_checkpoint(base_model_state_dict)
            return gr.Dropdown.update()

    def update_lora_model(self, lora_model_dropdown):
        lora_model_dropdown = os.path.join(self.personalized_model_dir, lora_model_dropdown)
        self.lora_model_state_dict = {}
        if lora_model_dropdown == "none": pass
        else:
            with safe_open(lora_model_dropdown, framework="pt", device="cpu") as f:
                for key in f.keys():
                    self.lora_model_state_dict[key] = f.get_tensor(key)
        return gr.Dropdown.update()

    def animate(
        self,
        stable_diffusion_dropdown,
        motion_module_dropdown,
        base_model_dropdown,
        lora_alpha_slider,
        prompt_textbox, 
        negative_prompt_textbox, 
        sampler_dropdown, 
        sample_step_slider, 
        width_slider, 
        length_slider, 
        height_slider, 
        cfg_scale_slider, 
        seed_textbox
    ):    
        if self.unet is None:
            raise gr.Error(f"Please select a pretrained model path.")
        if motion_module_dropdown == "": 
            raise gr.Error(f"Please select a motion module.")
        # if base_model_dropdown == "":
        #     raise gr.Error(f"Please select a base DreamBooth model.")

        if is_xformers_available(): self.unet.enable_xformers_memory_efficient_attention()

        pipeline = AnimationPipeline(
            vae=self.vae, text_encoder=self.text_encoder, tokenizer=self.tokenizer, unet=self.unet,
            scheduler=scheduler_dict[sampler_dropdown](**OmegaConf.to_container(self.inference_config.noise_scheduler_kwargs))
        ).to("cuda")
        
        if self.lora_model_state_dict != {}:
            print(f"Lora alpha: {lora_alpha_slider}")
            pipeline = convert_lora(copy.deepcopy(pipeline), self.lora_model_state_dict, alpha=lora_alpha_slider)
        pipeline.to("cuda")

        torch.cuda.empty_cache()
        
        seed_textbox = int(seed_textbox)
        if seed_textbox != -1 and seed_textbox != "": torch.manual_seed(seed_textbox)
        else: torch.seed()
        seed = torch.initial_seed()
        
        sample = pipeline(
            prompt_textbox,
            negative_prompt     = negative_prompt_textbox,
            num_inference_steps = sample_step_slider,
            guidance_scale      = cfg_scale_slider,
            width               = width_slider,
            height              = height_slider,
            video_length        = length_slider,
        ).videos

        save_sample_path = os.path.join(self.savedir_sample, f"{sample_idx}.mp4")
        save_videos_grid(sample, save_sample_path)
    
        sample_config = {
            "prompt": prompt_textbox,
            "n_prompt": negative_prompt_textbox,
            "sampler": sampler_dropdown,
            "num_inference_steps": sample_step_slider,
            "guidance_scale": cfg_scale_slider,
            "width": width_slider,
            "height": height_slider,
            "video_length": length_slider,
            "seed": seed
        }
        json_str = json.dumps(sample_config, indent=4)
        with open(os.path.join(self.savedir, "logs.json"), "a") as f:
            f.write(json_str)
            f.write("\n\n")
            
        return gr.Video.update(value=save_sample_path)
        

controller = AnimateController()


def ui():
    with gr.Blocks(css=css) as demo:
        gr.Markdown(
            """
            # [AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning](https://arxiv.org/abs/2307.04725)
            Yuwei Guo, Ceyuan Yang*, Anyi Rao, Yaohui Wang, Yu Qiao, Dahua Lin, Bo Dai (*Corresponding Author)<br>
            [Arxiv Report](https://arxiv.org/abs/2307.04725) | [Project Page](https://animatediff.github.io/) | [Github](https://github.com/guoyww/animatediff/)
            """
        )
        with gr.Column(variant="panel"):
            gr.Markdown(
                """
                ### 1. Model checkpoints (select pretrained model path first).
                """
            )
            with gr.Row():
                stable_diffusion_dropdown = gr.Dropdown(
                    label="Pretrained Model Path",
                    choices=controller.stable_diffusion_list,
                    interactive=True,
                )
                stable_diffusion_dropdown.change(fn=controller.update_stable_diffusion, inputs=[stable_diffusion_dropdown], outputs=[stable_diffusion_dropdown])
                
                stable_diffusion_refresh_button = gr.Button(value="\U0001F503", elem_classes="toolbutton")
                def update_stable_diffusion():
                    controller.refresh_stable_diffusion()
                    return gr.Dropdown.update(choices=controller.stable_diffusion_list)
                stable_diffusion_refresh_button.click(fn=update_stable_diffusion, inputs=[], outputs=[stable_diffusion_dropdown])

            with gr.Row():
                motion_module_dropdown = gr.Dropdown(
                    label="Select motion module",
                    choices=controller.motion_module_list,
                    interactive=True,
                )
                motion_module_dropdown.change(fn=controller.update_motion_module, inputs=[motion_module_dropdown], outputs=[motion_module_dropdown])
                
                motion_module_refresh_button = gr.Button(value="\U0001F503", elem_classes="toolbutton")
                def update_motion_module():
                    controller.refresh_motion_module()
                    return gr.Dropdown.update(choices=controller.motion_module_list)
                motion_module_refresh_button.click(fn=update_motion_module, inputs=[], outputs=[motion_module_dropdown])
                
                base_model_dropdown = gr.Dropdown(
                    label="Select base Dreambooth model (required)",
                    choices=controller.personalized_model_list,
                    interactive=True,
                )
                base_model_dropdown.change(fn=controller.update_base_model, inputs=[base_model_dropdown], outputs=[base_model_dropdown])
                
                lora_model_dropdown = gr.Dropdown(
                    label="Select LoRA model (optional)",
                    choices=["none"] + controller.personalized_model_list,
                    value="none",
                    interactive=True,
                )
                lora_model_dropdown.change(fn=controller.update_lora_model, inputs=[lora_model_dropdown], outputs=[lora_model_dropdown])
                
                lora_alpha_slider = gr.Slider(label="LoRA alpha", value=0.7, minimum=0, maximum=2, interactive=True)
                
                personalized_refresh_button = gr.Button(value="\U0001F503", elem_classes="toolbutton")
                def update_personalized_model():
                    controller.refresh_personalized_model()
                    return [
                        gr.Dropdown.update(choices=controller.personalized_model_list),
                        gr.Dropdown.update(choices=["none"] + controller.personalized_model_list)
                    ]
                personalized_refresh_button.click(fn=update_personalized_model, inputs=[], outputs=[base_model_dropdown, lora_model_dropdown])

        with gr.Column(variant="panel"):
            gr.Markdown(
                """
                ### 2. Configs for AnimateDiff.
                """
            )
            
            prompt_textbox = gr.Textbox(label="Prompt", lines=2)
            negative_prompt_textbox = gr.Textbox(label="Negative prompt", lines=2)
                
            with gr.Row().style(equal_height=False):
                with gr.Column():
                    with gr.Row():
                        sampler_dropdown   = gr.Dropdown(label="Sampling method", choices=list(scheduler_dict.keys()), value=list(scheduler_dict.keys())[0])
                        sample_step_slider = gr.Slider(label="Sampling steps", value=25, minimum=10, maximum=100, step=1)
                        
                    width_slider     = gr.Slider(label="Width",            value=512, minimum=256, maximum=1024, step=64)
                    height_slider    = gr.Slider(label="Height",           value=512, minimum=256, maximum=1024, step=64)
                    length_slider    = gr.Slider(label="Animation length", value=16,  minimum=8,   maximum=24,   step=1)
                    cfg_scale_slider = gr.Slider(label="CFG Scale",        value=7.5, minimum=0,   maximum=20)
                    
                    with gr.Row():
                        seed_textbox = gr.Textbox(label="Seed", value=-1)
                        seed_button  = gr.Button(value="\U0001F3B2", elem_classes="toolbutton")
                        seed_button.click(fn=lambda: gr.Textbox.update(value=random.randint(1, 1e8)), inputs=[], outputs=[seed_textbox])
            
                    generate_button = gr.Button(value="Generate", variant='primary')
                    
                result_video = gr.Video(label="Generated Animation", interactive=False)

            generate_button.click(
                fn=controller.animate,
                inputs=[
                    stable_diffusion_dropdown,
                    motion_module_dropdown,
                    base_model_dropdown,
                    lora_alpha_slider,
                    prompt_textbox, 
                    negative_prompt_textbox, 
                    sampler_dropdown, 
                    sample_step_slider, 
                    width_slider, 
                    length_slider, 
                    height_slider, 
                    cfg_scale_slider, 
                    seed_textbox,
                ],
                outputs=[result_video]
            )
            
    return demo


if __name__ == "__main__":
    demo = ui()
    demo.queue(max_size=20)
    demo.launch()


# import os
# import torch
# import random

# import gradio as gr
# from glob import glob
# from omegaconf import OmegaConf
# from safetensors import safe_open

# from diffusers import AutoencoderKL
# from diffusers import EulerDiscreteScheduler, DDIMScheduler
# from diffusers.utils.import_utils import is_xformers_available
# from transformers import CLIPTextModel, CLIPTokenizer

# from animatediff.models.unet import UNet3DConditionModel
# from animatediff.pipelines.pipeline_animation import AnimationPipeline
# from animatediff.utils.util import save_videos_grid
# from animatediff.utils.convert_from_ckpt import convert_ldm_unet_checkpoint, convert_ldm_clip_checkpoint, convert_ldm_vae_checkpoint


# pretrained_model_path = "models/StableDiffusion/stable-diffusion-v1-5"
# inference_config_path = "configs/inference/inference.yaml"

# css = """
# .toolbutton {
#     margin-buttom: 0em 0em 0em 0em;
#     max-width: 2.5em;
#     min-width: 2.5em !important;
#     height: 2.5em;
# }
# """

# examples = [
#     # 1-ToonYou
#     [
#         "toonyou_beta3.safetensors", 
#         "mm_sd_v14.ckpt", 
#         "masterpiece, best quality, 1girl, solo, cherry blossoms, hanami, pink flower, white flower, spring season, wisteria, petals, flower, plum blossoms, outdoors, falling petals, white hair, black eyes",
#         "worst quality, low quality, nsfw, logo",
#         512, 512, "13204175718326964000"
#     ],
#     # 2-Lyriel
#     [
#         "lyriel_v16.safetensors", 
#         "mm_sd_v15.ckpt", 
#         "A forbidden castle high up in the mountains, pixel art, intricate details2, hdr, intricate details, hyperdetailed5, natural skin texture, hyperrealism, soft light, sharp, game art, key visual, surreal",
#         "3d, cartoon, anime, sketches, worst quality, low quality, normal quality, lowres, normal quality, monochrome, grayscale, skin spots, acnes, skin blemishes, bad anatomy, girl, loli, young, large breasts, red eyes, muscular",
#         512, 512, "6681501646976930000"
#     ],
#     # 3-RCNZ
#     [
#         "rcnzCartoon3d_v10.safetensors", 
#         "mm_sd_v14.ckpt", 
#         "Jane Eyre with headphones, natural skin texture,4mm,k textures, soft cinematic light, adobe lightroom, photolab, hdr, intricate, elegant, highly detailed, sharp focus, cinematic look, soothing tones, insane details, intricate details, hyperdetailed, low contrast, soft cinematic light, dim colors, exposure blend, hdr, faded",
#         "deformed, distorted, disfigured, poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, mutated hands and fingers, disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation",
#         512, 512, "2416282124261060"
#     ],
#     # 4-MajicMix
#     [
#         "majicmixRealistic_v5Preview.safetensors", 
#         "mm_sd_v14.ckpt", 
#         "1girl, offshoulder, light smile, shiny skin best quality, masterpiece, photorealistic",
#         "bad hand, worst quality, low quality, normal quality, lowres, bad anatomy, bad hands, watermark, moles",
#         512, 512, "7132772652786303"
#     ],
#     # 5-RealisticVision
#     [
#         "realisticVisionV20_v20.safetensors", 
#         "mm_sd_v15.ckpt", 
#         "photo of coastline, rocks, storm weather, wind, waves, lightning, 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3",
#         "blur, haze, deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime, mutated hands and fingers, deformed, distorted, disfigured, poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, disconnected limbs, mutation, mutated, ugly, disgusting, amputation",
#         512, 512, "1490157606650685400"
#     ]
# ]

# # clean unrelated ckpts
# # ckpts = [
# #     "realisticVisionV40_v20Novae.safetensors",
# #     "majicmixRealistic_v5Preview.safetensors",
# #     "rcnzCartoon3d_v10.safetensors",
# #     "lyriel_v16.safetensors",
# #     "toonyou_beta3.safetensors"
# # ]

# # for path in glob(os.path.join("models", "DreamBooth_LoRA", "*.safetensors")):
# #     for ckpt in ckpts:
# #         if path.endswith(ckpt): break
# #     else:
# #         print(f"### Cleaning {path} ...")
# #         os.system(f"rm -rf {path}")

# # os.system(f"rm -rf {os.path.join('models', 'DreamBooth_LoRA', '*.safetensors')}")

# # os.system(f"bash download_bashscripts/1-ToonYou.sh")
# # os.system(f"bash download_bashscripts/2-Lyriel.sh")
# # os.system(f"bash download_bashscripts/3-RcnzCartoon.sh")
# # os.system(f"bash download_bashscripts/4-MajicMix.sh")
# # os.system(f"bash download_bashscripts/5-RealisticVision.sh")

# # clean Grdio cache
# print(f"### Cleaning cached examples ...")
# os.system(f"rm -rf gradio_cached_examples/")


# class AnimateController:
#     def __init__(self):
        
#         # config dirs
#         self.basedir                = os.getcwd()
#         self.stable_diffusion_dir   = os.path.join(self.basedir, "models", "StableDiffusion")
#         self.motion_module_dir      = os.path.join(self.basedir, "models", "Motion_Module")
#         self.personalized_model_dir = os.path.join(self.basedir, "models", "DreamBooth_LoRA")
#         self.savedir                = os.path.join(self.basedir, "samples")
#         os.makedirs(self.savedir, exist_ok=True)

#         self.base_model_list    = []
#         self.motion_module_list = []
        
#         self.selected_base_model    = None
#         self.selected_motion_module = None
        
#         self.refresh_motion_module()
#         self.refresh_personalized_model()
        
#         # config models
#         self.inference_config      = OmegaConf.load(inference_config_path)

#         self.tokenizer             = CLIPTokenizer.from_pretrained(pretrained_model_path, subfolder="tokenizer")
#         self.text_encoder          = CLIPTextModel.from_pretrained(pretrained_model_path, subfolder="text_encoder").cuda()
#         self.vae                   = AutoencoderKL.from_pretrained(pretrained_model_path, subfolder="vae").cuda()
#         self.unet                  = UNet3DConditionModel.from_pretrained_2d(pretrained_model_path, subfolder="unet", unet_additional_kwargs=OmegaConf.to_container(self.inference_config.unet_additional_kwargs)).cuda()
        
#         self.update_base_model(self.base_model_list[0])
#         self.update_motion_module(self.motion_module_list[0])
        
        
#     def refresh_motion_module(self):
#         motion_module_list = glob(os.path.join(self.motion_module_dir, "*.ckpt"))
#         self.motion_module_list = [os.path.basename(p) for p in motion_module_list]

#     def refresh_personalized_model(self):
#         base_model_list = glob(os.path.join(self.personalized_model_dir, "*.safetensors"))
#         self.base_model_list = [os.path.basename(p) for p in base_model_list]


#     def update_base_model(self, base_model_dropdown):
#         self.selected_base_model = base_model_dropdown
        
#         base_model_dropdown = os.path.join(self.personalized_model_dir, base_model_dropdown)
#         base_model_state_dict = {}
#         with safe_open(base_model_dropdown, framework="pt", device="cpu") as f:
#             for key in f.keys(): base_model_state_dict[key] = f.get_tensor(key)
                
#         converted_vae_checkpoint = convert_ldm_vae_checkpoint(base_model_state_dict, self.vae.config)
#         self.vae.load_state_dict(converted_vae_checkpoint)

#         converted_unet_checkpoint = convert_ldm_unet_checkpoint(base_model_state_dict, self.unet.config)
#         self.unet.load_state_dict(converted_unet_checkpoint, strict=False)

#         self.text_encoder = convert_ldm_clip_checkpoint(base_model_state_dict)
#         return gr.Dropdown.update()

#     def update_motion_module(self, motion_module_dropdown):
#         self.selected_motion_module = motion_module_dropdown
        
#         motion_module_dropdown = os.path.join(self.motion_module_dir, motion_module_dropdown)
#         motion_module_state_dict = torch.load(motion_module_dropdown, map_location="cpu")
#         _, unexpected = self.unet.load_state_dict(motion_module_state_dict, strict=False)
#         assert len(unexpected) == 0
#         return gr.Dropdown.update()
    
    
#     def animate(
#         self,
#         base_model_dropdown,
#         motion_module_dropdown,
#         prompt_textbox,
#         negative_prompt_textbox,
#         width_slider,
#         height_slider,
#         seed_textbox,
#     ):
#         if self.selected_base_model != base_model_dropdown: self.update_base_model(base_model_dropdown)
#         if self.selected_motion_module != motion_module_dropdown: self.update_motion_module(motion_module_dropdown)
        
#         if is_xformers_available(): self.unet.enable_xformers_memory_efficient_attention()

#         pipeline = AnimationPipeline(
#             vae=self.vae, text_encoder=self.text_encoder, tokenizer=self.tokenizer, unet=self.unet,
#             scheduler=DDIMScheduler(**OmegaConf.to_container(self.inference_config.noise_scheduler_kwargs))
#         ).to("cuda")
        
#         if int(seed_textbox) > 0: seed = int(seed_textbox)
#         else: seed = random.randint(1, 1e16)
#         torch.manual_seed(int(seed))
        
#         assert seed == torch.initial_seed()
#         print(f"### seed: {seed}")
        
#         generator = torch.Generator(device="cuda")
#         generator.manual_seed(seed)
        
#         sample = pipeline(
#             prompt_textbox,
#             negative_prompt     = negative_prompt_textbox,
#             num_inference_steps = 25,
#             guidance_scale      = 8.,
#             width               = width_slider,
#             height              = height_slider,
#             video_length        = 16,
#             generator           = generator,
#         ).videos

#         save_sample_path = os.path.join(self.savedir, f"sample.mp4")
#         save_videos_grid(sample, save_sample_path)
    
#         json_config = {
#             "prompt": prompt_textbox,
#             "n_prompt": negative_prompt_textbox,
#             "width": width_slider,
#             "height": height_slider,
#             "seed": seed,
#             "base_model": base_model_dropdown,
#             "motion_module": motion_module_dropdown,
#         }
#         return gr.Video.update(value=save_sample_path), gr.Json.update(value=json_config)
        

# controller = AnimateController()


# def ui():
#     with gr.Blocks(css=css) as demo:
#         gr.Markdown(
#             """
#             # AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning
#             Yuwei Guo, Ceyuan Yang*, Anyi Rao, Yaohui Wang, Yu Qiao, Dahua Lin, Bo Dai (*Corresponding Author)<br>
#             [Arxiv Report](https://arxiv.org/abs/2307.04725) | [Project Page](https://animatediff.github.io/) | [Github](https://github.com/guoyww/animatediff/)
#             """
#         )
#         gr.Markdown(
#             """
#             ### Quick Start
#             1. Select desired `Base DreamBooth Model`.
#             2. Select `Motion Module` from `mm_sd_v14.ckpt` and `mm_sd_v15.ckpt`. We recommend trying both of them for the best results.
#             3. Provide `Prompt` and `Negative Prompt` for each model. You are encouraged to refer to each model's webpage on CivitAI to learn how to write prompts for them. Below are the DreamBooth models in this demo. Click to visit their homepage.
#                 - [`toonyou_beta3.safetensors`](https://civitai.com/models/30240?modelVersionId=78775)
#                 - [`lyriel_v16.safetensors`](https://civitai.com/models/22922/lyriel)
#                 - [`rcnzCartoon3d_v10.safetensors`](https://civitai.com/models/66347?modelVersionId=71009)
#                 - [`majicmixRealistic_v5Preview.safetensors`](https://civitai.com/models/43331?modelVersionId=79068)
#                 - [`realisticVisionV20_v20.safetensors`](https://civitai.com/models/4201?modelVersionId=29460)
#             4. Click `Generate`, wait for ~1 min, and enjoy.
#             """
#         )
#         with gr.Row():
#             with gr.Column():
#                 base_model_dropdown     = gr.Dropdown( label="Base DreamBooth Model", choices=controller.base_model_list,    value=controller.base_model_list[0],    interactive=True )
#                 motion_module_dropdown  = gr.Dropdown( label="Motion Module",  choices=controller.motion_module_list, value=controller.motion_module_list[0], interactive=True )

#                 base_model_dropdown.change(fn=controller.update_base_model,       inputs=[base_model_dropdown],    outputs=[base_model_dropdown])
#                 motion_module_dropdown.change(fn=controller.update_motion_module, inputs=[motion_module_dropdown], outputs=[motion_module_dropdown])

#                 prompt_textbox          = gr.Textbox( label="Prompt",          lines=3 )
#                 negative_prompt_textbox = gr.Textbox( label="Negative Prompt", lines=3, value="worst quality, low quality, nsfw, logo")

#                 with gr.Accordion("Advance", open=False):
#                     with gr.Row():
#                         width_slider  = gr.Slider(  label="Width",  value=512, minimum=256, maximum=1024, step=64 )
#                         height_slider = gr.Slider(  label="Height", value=512, minimum=256, maximum=1024, step=64 )
#                     with gr.Row():
#                         seed_textbox = gr.Textbox( label="Seed",  value=-1)
#                         seed_button  = gr.Button(value="\U0001F3B2", elem_classes="toolbutton")
#                         seed_button.click(fn=lambda: gr.Textbox.update(value=random.randint(1, 1e16)), inputs=[], outputs=[seed_textbox])

#                 generate_button = gr.Button( value="Generate", variant='primary' )

#             with gr.Column():
#                 result_video = gr.Video( label="Generated Animation", interactive=False )
#                 json_config  = gr.Json( label="Config", value=None )

#             inputs  = [base_model_dropdown, motion_module_dropdown, prompt_textbox, negative_prompt_textbox, width_slider, height_slider, seed_textbox]
#             outputs = [result_video, json_config]
            
#             generate_button.click( fn=controller.animate, inputs=inputs, outputs=outputs )
                
#         gr.Examples( fn=controller.animate, examples=examples, inputs=inputs, outputs=outputs, cache_examples=True )
        
#     return demo


# if __name__ == "__main__":
#     demo = ui()
#     demo.queue(max_size=20)
#     demo.launch()