File size: 27,845 Bytes
4b30d84
 
 
 
 
 
1e66485
bc84695
4b30d84
 
 
 
 
 
 
70a03b7
6e97454
 
 
4b30d84
 
 
 
 
 
1e66485
4b30d84
 
77a9bc5
 
da043f2
4b30d84
 
06905b7
a7a1fa6
4b30d84
 
 
 
 
 
 
 
 
 
 
 
6e97454
 
4b30d84
 
 
 
 
6e97454
 
 
 
 
 
4b30d84
e1fa56f
4b30d84
 
 
 
 
 
 
6e97454
4b30d84
 
 
 
6e97454
4b30d84
 
 
 
6e97454
4b30d84
 
 
 
6e97454
4b30d84
 
 
 
 
 
 
 
 
 
ebb4814
4b30d84
 
 
6e97454
17146d5
6e97454
ebb4814
28c56c2
ebb4814
 
6e97454
28c56c2
6e97454
4b30d84
1e66485
28c56c2
1e66485
 
3c18f07
 
28c56c2
3c18f07
ebb4814
28c56c2
ebb4814
 
28c56c2
 
 
 
 
 
 
 
 
e1fa56f
 
 
 
 
 
28c56c2
 
 
ebb4814
 
 
 
28c56c2
 
 
 
ebb4814
 
 
 
28c56c2
ebb4814
8902214
 
 
ebb4814
 
1e66485
e1fa56f
4b30d84
 
 
 
 
 
 
 
1e66485
 
 
 
 
 
 
8902214
1e66485
 
 
 
 
 
 
4b30d84
8902214
 
 
 
 
 
4b30d84
 
1e66485
 
 
 
 
 
 
 
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e66485
8902214
6e97454
b4d3beb
6e97454
4b30d84
 
 
 
 
8902214
4b30d84
1e66485
4b30d84
 
 
 
 
6614219
1e66485
6e97454
1e66485
8902214
1e66485
 
4b30d84
1e66485
 
8902214
 
4b30d84
 
 
 
 
 
 
 
 
 
49ba457
 
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e66485
 
25d37d8
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9d68af7
 
 
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6e97454
 
fd46938
4b30d84
 
 
 
 
 
 
 
bc84695
4b30d84
 
 
70a03b7
4b30d84
 
 
 
 
6e97454
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fd46938
6e97454
4b30d84
 
6e97454
 
4b30d84
 
 
 
 
6e97454
 
 
4b30d84
 
 
 
6e97454
 
 
 
 
 
 
 
4b30d84
 
 
 
6e97454
 
 
 
 
 
 
 
4b30d84
 
 
 
3f8aed1
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6e97454
 
 
 
 
 
 
 
4b30d84
 
 
6e97454
 
 
 
 
 
 
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c0beaa0
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e66485
4b30d84
 
 
 
6477781
4b30d84
 
6477781
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6477781
4b30d84
6477781
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
97a8826
4b30d84
 
 
 
 
 
 
97a8826
4b30d84
 
 
 
 
 
 
 
 
 
 
6e97454
4b30d84
 
 
 
 
6e97454
4b30d84
6e97454
4b30d84
 
 
 
 
 
 
 
 
 
6e97454
4b30d84
6e97454
 
 
97a8826
4b30d84
 
6e97454
 
 
97a8826
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6e97454
 
 
 
 
4b30d84
 
 
 
 
 
 
1e66485
4b30d84
 
 
 
 
97a8826
4b30d84
 
 
 
6e97454
97a8826
4b30d84
 
 
 
 
97a8826
4b30d84
 
 
 
 
97a8826
4b30d84
6e97454
 
 
 
97a8826
6e97454
4b30d84
 
 
 
 
 
 
 
1e66485
6e97454
 
97a8826
6e97454
4b30d84
 
 
 
 
 
 
 
 
 
 
 
6e97454
4b30d84
97a8826
4b30d84
 
 
 
 
 
97a8826
4b30d84
 
 
 
 
97a8826
4b30d84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6e97454
4b30d84
 
 
 
 
 
 
1e66485
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
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
import random
import tempfile
import time
import gradio as gr
import numpy as np
import torch
import math
import re

from gradio import inputs
from diffusers import (
    AutoencoderKL,
    DDIMScheduler,
    UNet2DConditionModel,
)
from modules.model import (
    CrossAttnProcessor,
    StableDiffusionPipeline,
)
from torchvision import transforms
from transformers import CLIPTokenizer, CLIPTextModel
from PIL import Image
from pathlib import Path
from safetensors.torch import load_file
import modules.safe as _
from modules.lora import LoRANetwork

models = [
    ("AbyssOrangeMix2", "Korakoe/AbyssOrangeMix2-HF", 2),
    ("Pastal Mix", "andite/pastel-mix", 2),
    ("Basil Mix", "nuigurumi/basil_mix", 2)
]

keep_vram = ["Korakoe/AbyssOrangeMix2-HF", "andite/pastel-mix"]
base_name, base_model, clip_skip = models[0]

samplers_k_diffusion = [
    ("Euler a", "sample_euler_ancestral", {}),
    ("Euler", "sample_euler", {}),
    ("LMS", "sample_lms", {}),
    ("Heun", "sample_heun", {}),
    ("DPM2", "sample_dpm_2", {"discard_next_to_last_sigma": True}),
    ("DPM2 a", "sample_dpm_2_ancestral", {"discard_next_to_last_sigma": True}),
    ("DPM++ 2S a", "sample_dpmpp_2s_ancestral", {}),
    ("DPM++ 2M", "sample_dpmpp_2m", {}),
    ("DPM++ SDE", "sample_dpmpp_sde", {}),
    ("LMS Karras", "sample_lms", {"scheduler": "karras"}),
    ("DPM2 Karras", "sample_dpm_2", {"scheduler": "karras", "discard_next_to_last_sigma": True}),
    ("DPM2 a Karras", "sample_dpm_2_ancestral", {"scheduler": "karras", "discard_next_to_last_sigma": True}),
    ("DPM++ 2S a Karras", "sample_dpmpp_2s_ancestral", {"scheduler": "karras"}),
    ("DPM++ 2M Karras", "sample_dpmpp_2m", {"scheduler": "karras"}),
    ("DPM++ SDE Karras", "sample_dpmpp_sde", {"scheduler": "karras"}),
]

# samplers_diffusers = [
#     ("DDIMScheduler", "diffusers.schedulers.DDIMScheduler", {})
#     ("DDPMScheduler", "diffusers.schedulers.DDPMScheduler", {})
#     ("DEISMultistepScheduler", "diffusers.schedulers.DEISMultistepScheduler", {})
# ]

start_time = time.time()
timeout = 90

scheduler = DDIMScheduler.from_pretrained(
    base_model,
    subfolder="scheduler",
)
vae = AutoencoderKL.from_pretrained(
    "stabilityai/sd-vae-ft-ema", 
    torch_dtype=torch.float16
)
text_encoder = CLIPTextModel.from_pretrained(
    base_model,
    subfolder="text_encoder",
    torch_dtype=torch.float16,
)
tokenizer = CLIPTokenizer.from_pretrained(
    base_model,
    subfolder="tokenizer",
    torch_dtype=torch.float16,
)
unet = UNet2DConditionModel.from_pretrained(
    base_model,
    subfolder="unet",
    torch_dtype=torch.float16,
)
pipe = StableDiffusionPipeline(
    text_encoder=text_encoder,
    tokenizer=tokenizer,
    unet=unet,
    vae=vae,
    scheduler=scheduler,
)

unet.set_attn_processor(CrossAttnProcessor)
pipe.setup_text_encoder(clip_skip, text_encoder)
if torch.cuda.is_available():
    pipe = pipe.to("cuda")

def get_model_list():
    return models

te_cache = {
    base_model: text_encoder
}

unet_cache = {
    base_model: unet
}

lora_cache = {
    base_model: LoRANetwork(text_encoder, unet)
}

te_base_weight_length = text_encoder.get_input_embeddings().weight.data.shape[0]
original_prepare_for_tokenization = tokenizer.prepare_for_tokenization
current_model = base_model

def setup_model(name, lora_state=None, lora_scale=1.0):
    global pipe, current_model

    keys = [k[0] for k in models]
    model = models[keys.index(name)][1]
    if model not in unet_cache:
        unet = UNet2DConditionModel.from_pretrained(model, subfolder="unet", torch_dtype=torch.float16)
        text_encoder = CLIPTextModel.from_pretrained(model, subfolder="text_encoder", torch_dtype=torch.float16)

        unet_cache[model] = unet
        te_cache[model] = text_encoder
        lora_cache[model] = LoRANetwork(text_encoder, unet)

    if current_model != model:
        if current_model not in keep_vram:
            # offload current model
            unet_cache[current_model].to("cpu")
            te_cache[current_model].to("cpu")
            lora_cache[current_model].to("cpu")
        current_model = model

    local_te, local_unet, local_lora, = te_cache[model], unet_cache[model], lora_cache[model]
    local_unet.set_attn_processor(CrossAttnProcessor())
    local_lora.reset()
    clip_skip = models[keys.index(name)][2]

    if torch.cuda.is_available():
        local_unet.to("cuda")
        local_te.to("cuda")

    if lora_state is not None and lora_state != "":
        local_lora.load(lora_state, lora_scale)
        local_lora.to(local_unet.device, dtype=local_unet.dtype)

    pipe.text_encoder, pipe.unet = local_te, local_unet
    pipe.setup_unet(local_unet)
    pipe.tokenizer.prepare_for_tokenization = original_prepare_for_tokenization
    pipe.tokenizer.added_tokens_encoder = {}
    pipe.tokenizer.added_tokens_decoder = {}
    pipe.setup_text_encoder(clip_skip, local_te)
    return pipe


def error_str(error, title="Error"):
    return (
        f"""#### {title}
            {error}"""
        if error
        else ""
    )

def make_token_names(embs):
    all_tokens = []
    for name, vec in embs.items():
        tokens = [f'emb-{name}-{i}' for i in range(len(vec))]
        all_tokens.append(tokens)
    return all_tokens

def setup_tokenizer(tokenizer, embs):
    reg_match = [re.compile(fr"(?:^|(?<=\s|,)){k}(?=,|\s|$)") for k in embs.keys()]
    clip_keywords = [' '.join(s) for s in make_token_names(embs)]

    def parse_prompt(prompt: str):
        for m, v in zip(reg_match, clip_keywords):
            prompt = m.sub(v, prompt)
        return prompt

    def prepare_for_tokenization(self, text: str, is_split_into_words: bool = False, **kwargs):
        text = parse_prompt(text)
        r = original_prepare_for_tokenization(text, is_split_into_words, **kwargs)
        return r
        tokenizer.prepare_for_tokenization = prepare_for_tokenization.__get__(tokenizer, CLIPTokenizer)
    return [t for sublist in make_token_names(embs) for t in sublist]


def convert_size(size_bytes):
    if size_bytes == 0:
        return "0B"
    size_name = ("B", "KB", "MB", "GB", "TB", "PB", "EB", "ZB", "YB")
    i = int(math.floor(math.log(size_bytes, 1024)))
    p = math.pow(1024, i)
    s = round(size_bytes / p, 2)
    return "%s %s" % (s, size_name[i])

def inference(
    prompt,
    guidance,
    steps,
    width=512,
    height=512,
    seed=0,
    neg_prompt="",
    state=None,
    g_strength=0.4,
    img_input=None,
    i2i_scale=0.5,
    hr_enabled=False,
    hr_method="Latent",
    hr_scale=1.5,
    hr_denoise=0.8,
    sampler="DPM++ 2M Karras",
    embs=None,
    model=None,
    lora_state=None,
    lora_scale=None,
):
    if seed is None or seed == 0:
        seed = random.randint(0, 2147483647)

    pipe = setup_model(model, lora_state, lora_scale)
    generator = torch.Generator("cuda").manual_seed(int(seed))
    start_time = time.time()

    sampler_name, sampler_opt = None, None
    for label, funcname, options in samplers_k_diffusion:
        if label == sampler:
            sampler_name, sampler_opt = funcname, options

    tokenizer, text_encoder = pipe.tokenizer, pipe.text_encoder
    if embs is not None and len(embs) > 0:
        ti_embs = {}
        for name, file in embs.items():
            if str(file).endswith(".pt"):
                loaded_learned_embeds = torch.load(file, map_location="cpu")
            else:
                loaded_learned_embeds = load_file(file, device="cpu")
            loaded_learned_embeds = loaded_learned_embeds["string_to_param"]["*"] if "string_to_param" in loaded_learned_embeds else loaded_learned_embeds
            ti_embs[name] = loaded_learned_embeds

        if len(ti_embs) > 0:
            tokens = setup_tokenizer(tokenizer, ti_embs)
            added_tokens = tokenizer.add_tokens(tokens)
            delta_weight = torch.cat([val for val in ti_embs.values()], dim=0)

            assert added_tokens == delta_weight.shape[0]
            text_encoder.resize_token_embeddings(len(tokenizer))
            token_embeds = text_encoder.get_input_embeddings().weight.data
            token_embeds[-delta_weight.shape[0]:] = delta_weight

    config = {
        "negative_prompt": neg_prompt,
        "num_inference_steps": int(steps),
        "guidance_scale": guidance,
        "generator": generator,
        "sampler_name": sampler_name,
        "sampler_opt": sampler_opt,
        "pww_state": state,
        "pww_attn_weight": g_strength,
        "start_time": start_time,
        "timeout": timeout,
    }

    if img_input is not None:
        ratio = min(height / img_input.height, width / img_input.width)
        img_input = img_input.resize(
            (int(img_input.width * ratio), int(img_input.height * ratio)), Image.LANCZOS
        )
        result = pipe.img2img(prompt, image=img_input, strength=i2i_scale, **config)
    elif hr_enabled:
        result = pipe.txt2img(
            prompt,
            width=width,
            height=height,
            upscale=True,
            upscale_x=hr_scale,
            upscale_denoising_strength=hr_denoise,
            **config,
            **latent_upscale_modes[hr_method],
        )
    else:
        result = pipe.txt2img(prompt, width=width, height=height, **config)

    end_time = time.time()
    vram_free, vram_total = torch.cuda.mem_get_info()
    print(f"done: model={model}, res={width}x{height}, step={steps}, time={round(end_time-start_time, 2)}s, vram_alloc={convert_size(vram_total-vram_free)}/{convert_size(vram_total)}")
    return gr.Image.update(result[0][0], label=f"Initial Seed: {seed}")


color_list = []


def get_color(n):
    for _ in range(n - len(color_list)):
        color_list.append(tuple(np.random.random(size=3) * 256))
    return color_list


def create_mixed_img(current, state, w=512, h=512):
    w, h = int(w), int(h)
    image_np = np.full([h, w, 4], 255)
    if state is None:
        state = {}

    colors = get_color(len(state))
    idx = 0

    for key, item in state.items():
        if item["map"] is not None:
            m = item["map"] < 255
            alpha = 150
            if current == key:
                alpha = 200
            image_np[m] = colors[idx] + (alpha,)
        idx += 1

    return image_np


# width.change(apply_new_res, inputs=[width, height, global_stats], outputs=[global_stats, sp, rendered])
def apply_new_res(w, h, state):
    w, h = int(w), int(h)

    for key, item in state.items():
        if item["map"] is not None:
            item["map"] = resize(item["map"], w, h)

    update_img = gr.Image.update(value=create_mixed_img("", state, w, h))
    return state, update_img


def detect_text(text, state, width, height):
    
    if text is None or text == "":
        return None, None, gr.Radio.update(value=None), None

    t = text.split(",")
    new_state = {}

    for item in t:
        item = item.strip()
        if item == "":
            continue
        if state is not None and item in state:
            new_state[item] = {
                "map": state[item]["map"],
                "weight": state[item]["weight"],
                "mask_outsides": state[item]["mask_outsides"],
            }
        else:
            new_state[item] = {
                "map": None,
                "weight": 0.5,
                "mask_outsides": False
            }
    update = gr.Radio.update(choices=[key for key in new_state.keys()], value=None)
    update_img = gr.update(value=create_mixed_img("", new_state, width, height))
    update_sketch = gr.update(value=None, interactive=False)
    return new_state, update_sketch, update, update_img


def resize(img, w, h):
    trs = transforms.Compose(
        [
            transforms.ToPILImage(),
            transforms.Resize(min(h, w)),
            transforms.CenterCrop((h, w)),
        ]
    )
    result = np.array(trs(img), dtype=np.uint8)
    return result


def switch_canvas(entry, state, width, height):
    if entry == None:
        return None, 0.5, False, create_mixed_img("", state, width, height)

    return (
        gr.update(value=None, interactive=True),
        gr.update(value=state[entry]["weight"] if entry in state else 0.5),
        gr.update(value=state[entry]["mask_outsides"] if entry in state else False),
        create_mixed_img(entry, state, width, height),
    )


def apply_canvas(selected, draw, state, w, h):
    if selected in state:
        w, h = int(w), int(h)
        state[selected]["map"] = resize(draw, w, h)
    return state, gr.Image.update(value=create_mixed_img(selected, state, w, h))


def apply_weight(selected, weight, state):
    if selected in state:
        state[selected]["weight"] = weight
    return state


def apply_option(selected, mask, state):
    if selected in state:
        state[selected]["mask_outsides"] = mask
    return state


# sp2, radio, width, height, global_stats
def apply_image(image, selected, w, h, strgength, mask, state):
    if selected in state:
        state[selected] = {
            "map": resize(image, w, h), 
            "weight": strgength, 
            "mask_outsides": mask
        }
        
    return state, gr.Image.update(value=create_mixed_img(selected, state, w, h))


# [ti_state, lora_state, ti_vals, lora_vals, uploads]
def add_net(files, ti_state, lora_state):
    if files is None:
        return ti_state, "", lora_state, None

    for file in files:
        item = Path(file.name)
        stripedname = str(item.stem).strip()
        if item.suffix == ".pt":
            state_dict = torch.load(file.name, map_location="cpu")
        else:
            state_dict = load_file(file.name, device="cpu")
        if any("lora" in k for k in state_dict.keys()):
            lora_state = file.name
        else:
            ti_state[stripedname] = file.name

    return (
        ti_state,
        lora_state,
        gr.Text.update(f"{[key for key in ti_state.keys()]}"),
        gr.Text.update(f"{lora_state}"),
        gr.Files.update(value=None),
    )


# [ti_state, lora_state, ti_vals, lora_vals, uploads]
def clean_states(ti_state, lora_state):
    return (
        dict(),
        None,
        gr.Text.update(f""),
        gr.Text.update(f""),
        gr.File.update(value=None),
    )


latent_upscale_modes = {
    "Latent": {"upscale_method": "bilinear", "upscale_antialias": False},
    "Latent (antialiased)": {"upscale_method": "bilinear", "upscale_antialias": True},
    "Latent (bicubic)": {"upscale_method": "bicubic", "upscale_antialias": False},
    "Latent (bicubic antialiased)": {
        "upscale_method": "bicubic",
        "upscale_antialias": True,
    },
    "Latent (nearest)": {"upscale_method": "nearest", "upscale_antialias": False},
    "Latent (nearest-exact)": {
        "upscale_method": "nearest-exact",
        "upscale_antialias": False,
    },
}

css = """
.finetuned-diffusion-div div{
    display:inline-flex;
    align-items:center;
    gap:.8rem;
    font-size:1.75rem;
    padding-top:2rem;
}
.finetuned-diffusion-div div h1{
    font-weight:900;
    margin-bottom:7px
}
.finetuned-diffusion-div p{
    margin-bottom:10px;
    font-size:94%
}
.box {
  float: left;
  height: 20px;
  width: 20px;
  margin-bottom: 15px;
  border: 1px solid black;
  clear: both;
}
a{
    text-decoration:underline
}
.tabs{
    margin-top:0;
    margin-bottom:0
}
#gallery{
    min-height:20rem
}
.no-border {
    border: none !important;
}
 """
with gr.Blocks(css=css) as demo:
    gr.HTML(
        f"""
            <div class="finetuned-diffusion-div">
              <div>
                <h1>Demo for diffusion models</h1>
              </div>
              <p>Hso @ nyanko.sketch2img.gradio</p>
            </div>
        """
    )
    global_stats = gr.State(value={})

    with gr.Row():

        with gr.Column(scale=55):
            model = gr.Dropdown(
                choices=[k[0] for k in get_model_list()],
                label="Model",
                value=base_name,
            )
            image_out = gr.Image(height=512)
        # gallery = gr.Gallery(
        #     label="Generated images", show_label=False, elem_id="gallery"
        # ).style(grid=[1], height="auto")

        with gr.Column(scale=45):

            with gr.Group():

                with gr.Row():
                    with gr.Column(scale=70):

                        prompt = gr.Textbox(
                            label="Prompt",
                            value="loli cat girl, blue eyes, flat chest, solo, long messy silver hair, blue capelet, cat ears, cat tail, upper body",
                            show_label=True,
                            max_lines=4,
                            placeholder="Enter prompt.",
                        )
                        neg_prompt = gr.Textbox(
                            label="Negative Prompt",
                            value="bad quality, low quality, jpeg artifact, cropped",
                            show_label=True,
                            max_lines=4,
                            placeholder="Enter negative prompt.",
                        )

                    generate = gr.Button(value="Generate").style(
                        rounded=(False, True, True, False)
                    )

            with gr.Tab("Options"):

                with gr.Group():

                    # n_images = gr.Slider(label="Images", value=1, minimum=1, maximum=4, step=1)
                    with gr.Row():
                        guidance = gr.Slider(
                            label="Guidance scale", value=7.5, maximum=15
                        )
                        steps = gr.Slider(
                            label="Steps", value=25, minimum=2, maximum=50, step=1
                        )

                    with gr.Row():
                        width = gr.Slider(
                            label="Width", value=512, minimum=64, maximum=768, step=64
                        )
                        height = gr.Slider(
                            label="Height", value=512, minimum=64, maximum=768, step=64
                        )

                    sampler = gr.Dropdown(
                        value="DPM++ 2M Karras",
                        label="Sampler",
                        choices=[s[0] for s in samplers_k_diffusion],
                    )
                    seed = gr.Number(label="Seed (0 = random)", value=0)

            with gr.Tab("Image to image"):
                with gr.Group():

                    inf_image = gr.Image(
                        label="Image", height=256, tool="editor", type="pil"
                    )
                    inf_strength = gr.Slider(
                        label="Transformation strength",
                        minimum=0,
                        maximum=1,
                        step=0.01,
                        value=0.5,
                    )

            def res_cap(g, w, h, x):
                if g:
                    return f"Enable upscaler: {w}x{h} to {int(w*x)}x{int(h*x)}"
                else:
                    return "Enable upscaler"

            with gr.Tab("Hires fix"):
                with gr.Group():

                    hr_enabled = gr.Checkbox(label="Enable upscaler", value=False)
                    hr_method = gr.Dropdown(
                        [key for key in latent_upscale_modes.keys()],
                        value="Latent",
                        label="Upscale method",
                    )
                    hr_scale = gr.Slider(
                        label="Upscale factor",
                        minimum=1.0,
                        maximum=1.5,
                        step=0.1,
                        value=1.2,
                    )
                    hr_denoise = gr.Slider(
                        label="Denoising strength",
                        minimum=0.0,
                        maximum=1.0,
                        step=0.1,
                        value=0.8,
                    )

                    hr_scale.change(
                        lambda g, x, w, h: gr.Checkbox.update(
                            label=res_cap(g, w, h, x)
                        ),
                        inputs=[hr_enabled, hr_scale, width, height],
                        outputs=hr_enabled,
                        queue=False,
                    )
                    hr_enabled.change(
                        lambda g, x, w, h: gr.Checkbox.update(
                            label=res_cap(g, w, h, x)
                        ),
                        inputs=[hr_enabled, hr_scale, width, height],
                        outputs=hr_enabled,
                        queue=False,
                    )

            with gr.Tab("Embeddings/Loras"):

                ti_state = gr.State(dict())
                lora_state = gr.State()

                with gr.Group():
                    with gr.Row():
                        with gr.Column(scale=90):
                            ti_vals = gr.Text(label="Loaded embeddings")

                    with gr.Row():
                        with gr.Column(scale=90):
                            lora_vals = gr.Text(label="Loaded loras")

                with gr.Row():

                    uploads = gr.Files(label="Upload new embeddings/lora")

                    with gr.Column():
                        lora_scale = gr.Slider(
                            label="Lora scale",
                            minimum=0,
                            maximum=2,
                            step=0.01,
                            value=1.0,
                        )
                        btn = gr.Button(value="Upload")
                        btn_del = gr.Button(value="Reset")

                btn.click(
                    add_net,
                    inputs=[uploads, ti_state, lora_state],
                    outputs=[ti_state, lora_state, ti_vals, lora_vals, uploads],
                    queue=False,
                )
                btn_del.click(
                    clean_states,
                    inputs=[ti_state, lora_state],
                    outputs=[ti_state, lora_state, ti_vals, lora_vals, uploads],
                    queue=False,
                )

        # error_output = gr.Markdown()

    gr.HTML(
        f"""
            <div class="finetuned-diffusion-div">
              <div>
                <h1>Paint with words</h1>
              </div>
              <p>
                Will use the following formula: w = scale * token_weight_martix * log(1 + sigma) * max(qk).
              </p>
            </div>
        """
    )

    with gr.Row():

        with gr.Column(scale=55):

            rendered = gr.Image(
                invert_colors=True,
                source="canvas",
                interactive=False,
                image_mode="RGBA",
            )

        with gr.Column(scale=45):

            with gr.Group():
                with gr.Row():
                    with gr.Column(scale=70):
                        g_strength = gr.Slider(
                            label="Weight scaling",
                            minimum=0,
                            maximum=0.8,
                            step=0.01,
                            value=0.4,
                        )

                        text = gr.Textbox(
                            lines=2,
                            interactive=True,
                            label="Token to Draw: (Separate by comma)",
                        )

                        radio = gr.Radio([], label="Tokens")

                    sk_update = gr.Button(value="Update").style(
                        rounded=(False, True, True, False)
                    )

                # g_strength.change(lambda b: gr.update(f"Scaled additional attn: $w = {b} \log (1 + \sigma) \std (Q^T K)$."), inputs=g_strength, outputs=[g_output])

            with gr.Tab("SketchPad"):

                sp = gr.Image(
                    image_mode="L",
                    tool="sketch",
                    source="canvas",
                    interactive=False,
                )

                mask_outsides = gr.Checkbox(
                    label="Mask other areas", 
                    value=False
                )

                strength = gr.Slider(
                    label="Token strength",
                    minimum=0,
                    maximum=0.8,
                    step=0.01,
                    value=0.5,
                )


                sk_update.click(
                    detect_text,
                    inputs=[text, global_stats, width, height],
                    outputs=[global_stats, sp, radio, rendered],
                    queue=False,
                )
                radio.change(
                    switch_canvas,
                    inputs=[radio, global_stats, width, height],
                    outputs=[sp, strength, mask_outsides, rendered],
                    queue=False,
                )
                sp.edit(
                    apply_canvas,
                    inputs=[radio, sp, global_stats, width, height],
                    outputs=[global_stats, rendered],
                    queue=False,
                )
                strength.change(
                    apply_weight,
                    inputs=[radio, strength, global_stats],
                    outputs=[global_stats],
                    queue=False,
                )
                mask_outsides.change(
                    apply_option,
                    inputs=[radio, mask_outsides, global_stats],
                    outputs=[global_stats],
                    queue=False,
                )

            with gr.Tab("UploadFile"):

                sp2 = gr.Image(
                    image_mode="L",
                    source="upload",
                    shape=(512, 512),
                )

                mask_outsides2 = gr.Checkbox(
                    label="Mask other areas", 
                    value=False,
                )

                strength2 = gr.Slider(
                    label="Token strength",
                    minimum=0,
                    maximum=0.8,
                    step=0.01,
                    value=0.5,
                )

                apply_style = gr.Button(value="Apply")
                apply_style.click(
                    apply_image,
                    inputs=[sp2, radio, width, height, strength2, mask_outsides2, global_stats],
                    outputs=[global_stats, rendered],
                    queue=False,
                )

            width.change(
                apply_new_res,
                inputs=[width, height, global_stats],
                outputs=[global_stats, rendered],
                queue=False,
            )
            height.change(
                apply_new_res,
                inputs=[width, height, global_stats],
                outputs=[global_stats, rendered],
                queue=False,
            )

    # color_stats = gr.State(value={})
    # text.change(detect_color, inputs=[sp, text, color_stats], outputs=[color_stats, rendered])
    # sp.change(detect_color, inputs=[sp, text, color_stats], outputs=[color_stats, rendered])

    inputs = [
        prompt,
        guidance,
        steps,
        width,
        height,
        seed,
        neg_prompt,
        global_stats,
        g_strength,
        inf_image,
        inf_strength,
        hr_enabled,
        hr_method,
        hr_scale,
        hr_denoise,
        sampler,
        ti_state,
        model,
        lora_state,
        lora_scale,
    ]
    outputs = [image_out]
    prompt.submit(inference, inputs=inputs, outputs=outputs)
    generate.click(inference, inputs=inputs, outputs=outputs)

print(f"Space built in {time.time() - start_time:.2f} seconds")
# demo.launch(share=True)
demo.launch(enable_queue=True, server_name="0.0.0.0", server_port=7860)