File size: 32,974 Bytes
ac6acf2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from .k_diffusion import sampling as k_diffusion_sampling
from .extra_samplers import uni_pc
import torch
import collections
from comfy import model_management
import math
import logging
import comfy.sampler_helpers
import scipy
import numpy

def get_area_and_mult(conds, x_in, timestep_in):
    dims = tuple(x_in.shape[2:])
    area = None
    strength = 1.0

    if 'timestep_start' in conds:
        timestep_start = conds['timestep_start']
        if timestep_in[0] > timestep_start:
            return None
    if 'timestep_end' in conds:
        timestep_end = conds['timestep_end']
        if timestep_in[0] < timestep_end:
            return None
    if 'area' in conds:
        area = list(conds['area'])
    if 'strength' in conds:
        strength = conds['strength']

    input_x = x_in
    if area is not None:
        for i in range(len(dims)):
            area[i] = min(input_x.shape[i + 2] - area[len(dims) + i], area[i])
            input_x = input_x.narrow(i + 2, area[len(dims) + i], area[i])

    if 'mask' in conds:
        # Scale the mask to the size of the input
        # The mask should have been resized as we began the sampling process
        mask_strength = 1.0
        if "mask_strength" in conds:
            mask_strength = conds["mask_strength"]
        mask = conds['mask']
        assert(mask.shape[1:] == x_in.shape[2:])

        mask = mask[:input_x.shape[0]]
        if area is not None:
            for i in range(len(dims)):
                mask = mask.narrow(i + 1, area[len(dims) + i], area[i])

        mask = mask * mask_strength
        mask = mask.unsqueeze(1).repeat(input_x.shape[0] // mask.shape[0], input_x.shape[1], 1, 1)
    else:
        mask = torch.ones_like(input_x)
    mult = mask * strength

    if 'mask' not in conds and area is not None:
        rr = 8
        for i in range(len(dims)):
            if area[len(dims) + i] != 0:
                for t in range(rr):
                    m = mult.narrow(i + 2, t, 1)
                    m *= ((1.0/rr) * (t + 1))
            if (area[i] + area[len(dims) + i]) < x_in.shape[i + 2]:
                for t in range(rr):
                    m = mult.narrow(i + 2, area[i] - 1 - t, 1)
                    m *= ((1.0/rr) * (t + 1))

    conditioning = {}
    model_conds = conds["model_conds"]
    for c in model_conds:
        conditioning[c] = model_conds[c].process_cond(batch_size=x_in.shape[0], device=x_in.device, area=area)

    control = conds.get('control', None)

    patches = None
    if 'gligen' in conds:
        gligen = conds['gligen']
        patches = {}
        gligen_type = gligen[0]
        gligen_model = gligen[1]
        if gligen_type == "position":
            gligen_patch = gligen_model.model.set_position(input_x.shape, gligen[2], input_x.device)
        else:
            gligen_patch = gligen_model.model.set_empty(input_x.shape, input_x.device)

        patches['middle_patch'] = [gligen_patch]

    cond_obj = collections.namedtuple('cond_obj', ['input_x', 'mult', 'conditioning', 'area', 'control', 'patches'])
    return cond_obj(input_x, mult, conditioning, area, control, patches)

def cond_equal_size(c1, c2):
    if c1 is c2:
        return True
    if c1.keys() != c2.keys():
        return False
    for k in c1:
        if not c1[k].can_concat(c2[k]):
            return False
    return True

def can_concat_cond(c1, c2):
    if c1.input_x.shape != c2.input_x.shape:
        return False

    def objects_concatable(obj1, obj2):
        if (obj1 is None) != (obj2 is None):
            return False
        if obj1 is not None:
            if obj1 is not obj2:
                return False
        return True

    if not objects_concatable(c1.control, c2.control):
        return False

    if not objects_concatable(c1.patches, c2.patches):
        return False

    return cond_equal_size(c1.conditioning, c2.conditioning)

def cond_cat(c_list):
    c_crossattn = []
    c_concat = []
    c_adm = []
    crossattn_max_len = 0

    temp = {}
    for x in c_list:
        for k in x:
            cur = temp.get(k, [])
            cur.append(x[k])
            temp[k] = cur

    out = {}
    for k in temp:
        conds = temp[k]
        out[k] = conds[0].concat(conds[1:])

    return out

def calc_cond_batch(model, conds, x_in, timestep, model_options):
    out_conds = []
    out_counts = []
    to_run = []

    for i in range(len(conds)):
        out_conds.append(torch.zeros_like(x_in))
        out_counts.append(torch.ones_like(x_in) * 1e-37)

        cond = conds[i]
        if cond is not None:
            for x in cond:
                p = get_area_and_mult(x, x_in, timestep)
                if p is None:
                    continue

                to_run += [(p, i)]

    while len(to_run) > 0:
        first = to_run[0]
        first_shape = first[0][0].shape
        to_batch_temp = []
        for x in range(len(to_run)):
            if can_concat_cond(to_run[x][0], first[0]):
                to_batch_temp += [x]

        to_batch_temp.reverse()
        to_batch = to_batch_temp[:1]

        free_memory = model_management.get_free_memory(x_in.device)
        for i in range(1, len(to_batch_temp) + 1):
            batch_amount = to_batch_temp[:len(to_batch_temp)//i]
            input_shape = [len(batch_amount) * first_shape[0]] + list(first_shape)[1:]
            if model.memory_required(input_shape) < free_memory:
                to_batch = batch_amount
                break

        input_x = []
        mult = []
        c = []
        cond_or_uncond = []
        area = []
        control = None
        patches = None
        for x in to_batch:
            o = to_run.pop(x)
            p = o[0]
            input_x.append(p.input_x)
            mult.append(p.mult)
            c.append(p.conditioning)
            area.append(p.area)
            cond_or_uncond.append(o[1])
            control = p.control
            patches = p.patches

        batch_chunks = len(cond_or_uncond)
        input_x = torch.cat(input_x)
        c = cond_cat(c)
        timestep_ = torch.cat([timestep] * batch_chunks)

        if control is not None:
            c['control'] = control.get_control(input_x, timestep_, c, len(cond_or_uncond))

        transformer_options = {}
        if 'transformer_options' in model_options:
            transformer_options = model_options['transformer_options'].copy()

        if patches is not None:
            if "patches" in transformer_options:
                cur_patches = transformer_options["patches"].copy()
                for p in patches:
                    if p in cur_patches:
                        cur_patches[p] = cur_patches[p] + patches[p]
                    else:
                        cur_patches[p] = patches[p]
                transformer_options["patches"] = cur_patches
            else:
                transformer_options["patches"] = patches

        transformer_options["cond_or_uncond"] = cond_or_uncond[:]
        transformer_options["sigmas"] = timestep

        c['transformer_options'] = transformer_options

        if 'model_function_wrapper' in model_options:
            output = model_options['model_function_wrapper'](model.apply_model, {"input": input_x, "timestep": timestep_, "c": c, "cond_or_uncond": cond_or_uncond}).chunk(batch_chunks)
        else:
            output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)

        for o in range(batch_chunks):
            cond_index = cond_or_uncond[o]
            a = area[o]
            if a is None:
                out_conds[cond_index] += output[o] * mult[o]
                out_counts[cond_index] += mult[o]
            else:
                out_c = out_conds[cond_index]
                out_cts = out_counts[cond_index]
                dims = len(a) // 2
                for i in range(dims):
                    out_c = out_c.narrow(i + 2, a[i + dims], a[i])
                    out_cts = out_cts.narrow(i + 2, a[i + dims], a[i])
                out_c += output[o] * mult[o]
                out_cts += mult[o]

    for i in range(len(out_conds)):
        out_conds[i] /= out_counts[i]

    return out_conds

def calc_cond_uncond_batch(model, cond, uncond, x_in, timestep, model_options): #TODO: remove
    logging.warning("WARNING: The comfy.samplers.calc_cond_uncond_batch function is deprecated please use the calc_cond_batch one instead.")
    return tuple(calc_cond_batch(model, [cond, uncond], x_in, timestep, model_options))

def cfg_function(model, cond_pred, uncond_pred, cond_scale, x, timestep, model_options={}, cond=None, uncond=None):
    if "sampler_cfg_function" in model_options:
        args = {"cond": x - cond_pred, "uncond": x - uncond_pred, "cond_scale": cond_scale, "timestep": timestep, "input": x, "sigma": timestep,
                "cond_denoised": cond_pred, "uncond_denoised": uncond_pred, "model": model, "model_options": model_options}
        cfg_result = x - model_options["sampler_cfg_function"](args)
    else:
        cfg_result = uncond_pred + (cond_pred - uncond_pred) * cond_scale

    for fn in model_options.get("sampler_post_cfg_function", []):
        args = {"denoised": cfg_result, "cond": cond, "uncond": uncond, "model": model, "uncond_denoised": uncond_pred, "cond_denoised": cond_pred,
                "sigma": timestep, "model_options": model_options, "input": x}
        cfg_result = fn(args)

    return cfg_result

#The main sampling function shared by all the samplers
#Returns denoised
def sampling_function(model, x, timestep, uncond, cond, cond_scale, model_options={}, seed=None):
    if math.isclose(cond_scale, 1.0) and model_options.get("disable_cfg1_optimization", False) == False:
        uncond_ = None
    else:
        uncond_ = uncond

    conds = [cond, uncond_]
    out = calc_cond_batch(model, conds, x, timestep, model_options)

    for fn in model_options.get("sampler_pre_cfg_function", []):
        args = {"conds":conds, "conds_out": out, "cond_scale": cond_scale, "timestep": timestep,
                "input": x, "sigma": timestep, "model": model, "model_options": model_options}
        out  = fn(args)

    return cfg_function(model, out[0], out[1], cond_scale, x, timestep, model_options=model_options, cond=cond, uncond=uncond_)


class KSamplerX0Inpaint:
    def __init__(self, model, sigmas):
        self.inner_model = model
        self.sigmas = sigmas
    def __call__(self, x, sigma, denoise_mask, model_options={}, seed=None):
        if denoise_mask is not None:
            if "denoise_mask_function" in model_options:
                denoise_mask = model_options["denoise_mask_function"](sigma, denoise_mask, extra_options={"model": self.inner_model, "sigmas": self.sigmas})
            latent_mask = 1. - denoise_mask
            x = x * denoise_mask + self.inner_model.inner_model.model_sampling.noise_scaling(sigma.reshape([sigma.shape[0]] + [1] * (len(self.noise.shape) - 1)), self.noise, self.latent_image) * latent_mask
        out = self.inner_model(x, sigma, model_options=model_options, seed=seed)
        if denoise_mask is not None:
            out = out * denoise_mask + self.latent_image * latent_mask
        return out

def simple_scheduler(model_sampling, steps):
    s = model_sampling
    sigs = []
    ss = len(s.sigmas) / steps
    for x in range(steps):
        sigs += [float(s.sigmas[-(1 + int(x * ss))])]
    sigs += [0.0]
    return torch.FloatTensor(sigs)

def ddim_scheduler(model_sampling, steps):
    s = model_sampling
    sigs = []
    x = 1
    if math.isclose(float(s.sigmas[x]), 0, abs_tol=0.00001):
        steps += 1
        sigs = []
    else:
        sigs = [0.0]

    ss = max(len(s.sigmas) // steps, 1)
    while x < len(s.sigmas):
        sigs += [float(s.sigmas[x])]
        x += ss
    sigs = sigs[::-1]
    return torch.FloatTensor(sigs)

def normal_scheduler(model_sampling, steps, sgm=False, floor=False):
    s = model_sampling
    start = s.timestep(s.sigma_max)
    end = s.timestep(s.sigma_min)

    append_zero = True
    if sgm:
        timesteps = torch.linspace(start, end, steps + 1)[:-1]
    else:
        if math.isclose(float(s.sigma(end)), 0, abs_tol=0.00001):
            steps += 1
            append_zero = False
        timesteps = torch.linspace(start, end, steps)

    sigs = []
    for x in range(len(timesteps)):
        ts = timesteps[x]
        sigs.append(float(s.sigma(ts)))

    if append_zero:
        sigs += [0.0]

    return torch.FloatTensor(sigs)

# Implemented based on: https://arxiv.org/abs/2407.12173
def beta_scheduler(model_sampling, steps, alpha=0.6, beta=0.6):
    total_timesteps = (len(model_sampling.sigmas) - 1)
    ts = 1 - numpy.linspace(0, 1, steps, endpoint=False)
    ts = numpy.rint(scipy.stats.beta.ppf(ts, alpha, beta) * total_timesteps)

    sigs = []
    for t in ts:
        sigs += [float(model_sampling.sigmas[int(t)])]
    sigs += [0.0]
    return torch.FloatTensor(sigs)

def get_mask_aabb(masks):
    if masks.numel() == 0:
        return torch.zeros((0, 4), device=masks.device, dtype=torch.int)

    b = masks.shape[0]

    bounding_boxes = torch.zeros((b, 4), device=masks.device, dtype=torch.int)
    is_empty = torch.zeros((b), device=masks.device, dtype=torch.bool)
    for i in range(b):
        mask = masks[i]
        if mask.numel() == 0:
            continue
        if torch.max(mask != 0) == False:
            is_empty[i] = True
            continue
        y, x = torch.where(mask)
        bounding_boxes[i, 0] = torch.min(x)
        bounding_boxes[i, 1] = torch.min(y)
        bounding_boxes[i, 2] = torch.max(x)
        bounding_boxes[i, 3] = torch.max(y)

    return bounding_boxes, is_empty

def resolve_areas_and_cond_masks_multidim(conditions, dims, device):
    # We need to decide on an area outside the sampling loop in order to properly generate opposite areas of equal sizes.
    # While we're doing this, we can also resolve the mask device and scaling for performance reasons
    for i in range(len(conditions)):
        c = conditions[i]
        if 'area' in c:
            area = c['area']
            if area[0] == "percentage":
                modified = c.copy()
                a = area[1:]
                a_len = len(a) // 2
                area = ()
                for d in range(len(dims)):
                    area += (max(1, round(a[d] * dims[d])),)
                for d in range(len(dims)):
                    area += (round(a[d + a_len] * dims[d]),)

                modified['area'] = area
                c = modified
                conditions[i] = c

        if 'mask' in c:
            mask = c['mask']
            mask = mask.to(device=device)
            modified = c.copy()
            if len(mask.shape) == len(dims):
                mask = mask.unsqueeze(0)
            if mask.shape[1:] != dims:
                mask = torch.nn.functional.interpolate(mask.unsqueeze(1), size=dims, mode='bilinear', align_corners=False).squeeze(1)

            if modified.get("set_area_to_bounds", False): #TODO: handle dim != 2
                bounds = torch.max(torch.abs(mask),dim=0).values.unsqueeze(0)
                boxes, is_empty = get_mask_aabb(bounds)
                if is_empty[0]:
                    # Use the minimum possible size for efficiency reasons. (Since the mask is all-0, this becomes a noop anyway)
                    modified['area'] = (8, 8, 0, 0)
                else:
                    box = boxes[0]
                    H, W, Y, X = (box[3] - box[1] + 1, box[2] - box[0] + 1, box[1], box[0])
                    H = max(8, H)
                    W = max(8, W)
                    area = (int(H), int(W), int(Y), int(X))
                    modified['area'] = area

            modified['mask'] = mask
            conditions[i] = modified

def resolve_areas_and_cond_masks(conditions, h, w, device):
    logging.warning("WARNING: The comfy.samplers.resolve_areas_and_cond_masks function is deprecated please use the resolve_areas_and_cond_masks_multidim one instead.")
    return resolve_areas_and_cond_masks_multidim(conditions, [h, w], device)

def create_cond_with_same_area_if_none(conds, c): #TODO: handle dim != 2
    if 'area' not in c:
        return

    c_area = c['area']
    smallest = None
    for x in conds:
        if 'area' in x:
            a = x['area']
            if c_area[2] >= a[2] and c_area[3] >= a[3]:
                if a[0] + a[2] >= c_area[0] + c_area[2]:
                    if a[1] + a[3] >= c_area[1] + c_area[3]:
                        if smallest is None:
                            smallest = x
                        elif 'area' not in smallest:
                            smallest = x
                        else:
                            if smallest['area'][0] * smallest['area'][1] > a[0] * a[1]:
                                smallest = x
        else:
            if smallest is None:
                smallest = x
    if smallest is None:
        return
    if 'area' in smallest:
        if smallest['area'] == c_area:
            return

    out = c.copy()
    out['model_conds'] = smallest['model_conds'].copy() #TODO: which fields should be copied?
    conds += [out]

def calculate_start_end_timesteps(model, conds):
    s = model.model_sampling
    for t in range(len(conds)):
        x = conds[t]

        timestep_start = None
        timestep_end = None
        if 'start_percent' in x:
            timestep_start = s.percent_to_sigma(x['start_percent'])
        if 'end_percent' in x:
            timestep_end = s.percent_to_sigma(x['end_percent'])

        if (timestep_start is not None) or (timestep_end is not None):
            n = x.copy()
            if (timestep_start is not None):
                n['timestep_start'] = timestep_start
            if (timestep_end is not None):
                n['timestep_end'] = timestep_end
            conds[t] = n

def pre_run_control(model, conds):
    s = model.model_sampling
    for t in range(len(conds)):
        x = conds[t]

        timestep_start = None
        timestep_end = None
        percent_to_timestep_function = lambda a: s.percent_to_sigma(a)
        if 'control' in x:
            x['control'].pre_run(model, percent_to_timestep_function)

def apply_empty_x_to_equal_area(conds, uncond, name, uncond_fill_func):
    cond_cnets = []
    cond_other = []
    uncond_cnets = []
    uncond_other = []
    for t in range(len(conds)):
        x = conds[t]
        if 'area' not in x:
            if name in x and x[name] is not None:
                cond_cnets.append(x[name])
            else:
                cond_other.append((x, t))
    for t in range(len(uncond)):
        x = uncond[t]
        if 'area' not in x:
            if name in x and x[name] is not None:
                uncond_cnets.append(x[name])
            else:
                uncond_other.append((x, t))

    if len(uncond_cnets) > 0:
        return

    for x in range(len(cond_cnets)):
        temp = uncond_other[x % len(uncond_other)]
        o = temp[0]
        if name in o and o[name] is not None:
            n = o.copy()
            n[name] = uncond_fill_func(cond_cnets, x)
            uncond += [n]
        else:
            n = o.copy()
            n[name] = uncond_fill_func(cond_cnets, x)
            uncond[temp[1]] = n

def encode_model_conds(model_function, conds, noise, device, prompt_type, **kwargs):
    for t in range(len(conds)):
        x = conds[t]
        params = x.copy()
        params["device"] = device
        params["noise"] = noise
        default_width = None
        if len(noise.shape) >= 4: #TODO: 8 multiple should be set by the model
            default_width = noise.shape[3] * 8
        params["width"] = params.get("width", default_width)
        params["height"] = params.get("height", noise.shape[2] * 8)
        params["prompt_type"] = params.get("prompt_type", prompt_type)
        for k in kwargs:
            if k not in params:
                params[k] = kwargs[k]

        out = model_function(**params)
        x = x.copy()
        model_conds = x['model_conds'].copy()
        for k in out:
            model_conds[k] = out[k]
        x['model_conds'] = model_conds
        conds[t] = x
    return conds

class Sampler:
    def sample(self):
        pass

    def max_denoise(self, model_wrap, sigmas):
        max_sigma = float(model_wrap.inner_model.model_sampling.sigma_max)
        sigma = float(sigmas[0])
        return math.isclose(max_sigma, sigma, rel_tol=1e-05) or sigma > max_sigma

KSAMPLER_NAMES = ["euler", "euler_cfg_pp", "euler_ancestral", "euler_ancestral_cfg_pp", "heun", "heunpp2","dpm_2", "dpm_2_ancestral",
                  "lms", "dpm_fast", "dpm_adaptive", "dpmpp_2s_ancestral", "dpmpp_sde", "dpmpp_sde_gpu",
                  "dpmpp_2m", "dpmpp_2m_sde", "dpmpp_2m_sde_gpu", "dpmpp_3m_sde", "dpmpp_3m_sde_gpu", "ddpm", "lcm",
                  "ipndm", "ipndm_v", "deis"]

class KSAMPLER(Sampler):
    def __init__(self, sampler_function, extra_options={}, inpaint_options={}):
        self.sampler_function = sampler_function
        self.extra_options = extra_options
        self.inpaint_options = inpaint_options

    def sample(self, model_wrap, sigmas, extra_args, callback, noise, latent_image=None, denoise_mask=None, disable_pbar=False):
        extra_args["denoise_mask"] = denoise_mask
        model_k = KSamplerX0Inpaint(model_wrap, sigmas)
        model_k.latent_image = latent_image
        if self.inpaint_options.get("random", False): #TODO: Should this be the default?
            generator = torch.manual_seed(extra_args.get("seed", 41) + 1)
            model_k.noise = torch.randn(noise.shape, generator=generator, device="cpu").to(noise.dtype).to(noise.device)
        else:
            model_k.noise = noise

        noise = model_wrap.inner_model.model_sampling.noise_scaling(sigmas[0], noise, latent_image, self.max_denoise(model_wrap, sigmas))

        k_callback = None
        total_steps = len(sigmas) - 1
        if callback is not None:
            k_callback = lambda x: callback(x["i"], x["denoised"], x["x"], total_steps)

        samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, **self.extra_options)
        samples = model_wrap.inner_model.model_sampling.inverse_noise_scaling(sigmas[-1], samples)
        return samples


def ksampler(sampler_name, extra_options={}, inpaint_options={}):
    if sampler_name == "dpm_fast":
        def dpm_fast_function(model, noise, sigmas, extra_args, callback, disable):
            if len(sigmas) <= 1:
                return noise

            sigma_min = sigmas[-1]
            if sigma_min == 0:
                sigma_min = sigmas[-2]
            total_steps = len(sigmas) - 1
            return k_diffusion_sampling.sample_dpm_fast(model, noise, sigma_min, sigmas[0], total_steps, extra_args=extra_args, callback=callback, disable=disable)
        sampler_function = dpm_fast_function
    elif sampler_name == "dpm_adaptive":
        def dpm_adaptive_function(model, noise, sigmas, extra_args, callback, disable, **extra_options):
            if len(sigmas) <= 1:
                return noise

            sigma_min = sigmas[-1]
            if sigma_min == 0:
                sigma_min = sigmas[-2]
            return k_diffusion_sampling.sample_dpm_adaptive(model, noise, sigma_min, sigmas[0], extra_args=extra_args, callback=callback, disable=disable, **extra_options)
        sampler_function = dpm_adaptive_function
    else:
        sampler_function = getattr(k_diffusion_sampling, "sample_{}".format(sampler_name))

    return KSAMPLER(sampler_function, extra_options, inpaint_options)


def process_conds(model, noise, conds, device, latent_image=None, denoise_mask=None, seed=None):
    for k in conds:
        conds[k] = conds[k][:]
        resolve_areas_and_cond_masks_multidim(conds[k], noise.shape[2:], device)

    for k in conds:
        calculate_start_end_timesteps(model, conds[k])

    if hasattr(model, 'extra_conds'):
        for k in conds:
            conds[k] = encode_model_conds(model.extra_conds, conds[k], noise, device, k, latent_image=latent_image, denoise_mask=denoise_mask, seed=seed)

    #make sure each cond area has an opposite one with the same area
    for k in conds:
        for c in conds[k]:
            for kk in conds:
                if k != kk:
                    create_cond_with_same_area_if_none(conds[kk], c)

    for k in conds:
        pre_run_control(model, conds[k])

    if "positive" in conds:
        positive = conds["positive"]
        for k in conds:
            if k != "positive":
                apply_empty_x_to_equal_area(list(filter(lambda c: c.get('control_apply_to_uncond', False) == True, positive)), conds[k], 'control', lambda cond_cnets, x: cond_cnets[x])
                apply_empty_x_to_equal_area(positive, conds[k], 'gligen', lambda cond_cnets, x: cond_cnets[x])

    return conds

class CFGGuider:
    def __init__(self, model_patcher):
        self.model_patcher = model_patcher
        self.model_options = model_patcher.model_options
        self.original_conds = {}
        self.cfg = 1.0

    def set_conds(self, positive, negative):
        self.inner_set_conds({"positive": positive, "negative": negative})

    def set_cfg(self, cfg):
        self.cfg = cfg

    def inner_set_conds(self, conds):
        for k in conds:
            self.original_conds[k] = comfy.sampler_helpers.convert_cond(conds[k])

    def __call__(self, *args, **kwargs):
        return self.predict_noise(*args, **kwargs)

    def predict_noise(self, x, timestep, model_options={}, seed=None):
        return sampling_function(self.inner_model, x, timestep, self.conds.get("negative", None), self.conds.get("positive", None), self.cfg, model_options=model_options, seed=seed)

    def inner_sample(self, noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed):
        if latent_image is not None and torch.count_nonzero(latent_image) > 0: #Don't shift the empty latent image.
            latent_image = self.inner_model.process_latent_in(latent_image)

        self.conds = process_conds(self.inner_model, noise, self.conds, device, latent_image, denoise_mask, seed)

        extra_args = {"model_options": self.model_options, "seed":seed}

        samples = sampler.sample(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar)
        return self.inner_model.process_latent_out(samples.to(torch.float32))

    def sample(self, noise, latent_image, sampler, sigmas, denoise_mask=None, callback=None, disable_pbar=False, seed=None):
        if sigmas.shape[-1] == 0:
            return latent_image

        self.conds = {}
        for k in self.original_conds:
            self.conds[k] = list(map(lambda a: a.copy(), self.original_conds[k]))

        self.inner_model, self.conds, self.loaded_models = comfy.sampler_helpers.prepare_sampling(self.model_patcher, noise.shape, self.conds)
        device = self.model_patcher.load_device

        if denoise_mask is not None:
            denoise_mask = comfy.sampler_helpers.prepare_mask(denoise_mask, noise.shape, device)

        noise = noise.to(device)
        latent_image = latent_image.to(device)
        sigmas = sigmas.to(device)

        output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)

        comfy.sampler_helpers.cleanup_models(self.conds, self.loaded_models)
        del self.inner_model
        del self.conds
        del self.loaded_models
        return output


def sample(model, noise, positive, negative, cfg, device, sampler, sigmas, model_options={}, latent_image=None, denoise_mask=None, callback=None, disable_pbar=False, seed=None):
    cfg_guider = CFGGuider(model)
    cfg_guider.set_conds(positive, negative)
    cfg_guider.set_cfg(cfg)
    return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)


SCHEDULER_NAMES = ["normal", "karras", "exponential", "sgm_uniform", "simple", "ddim_uniform", "beta"]
SAMPLER_NAMES = KSAMPLER_NAMES + ["ddim", "uni_pc", "uni_pc_bh2"]

def calculate_sigmas(model_sampling, scheduler_name, steps):
    if scheduler_name == "karras":
        sigmas = k_diffusion_sampling.get_sigmas_karras(n=steps, sigma_min=float(model_sampling.sigma_min), sigma_max=float(model_sampling.sigma_max))
    elif scheduler_name == "exponential":
        sigmas = k_diffusion_sampling.get_sigmas_exponential(n=steps, sigma_min=float(model_sampling.sigma_min), sigma_max=float(model_sampling.sigma_max))
    elif scheduler_name == "normal":
        sigmas = normal_scheduler(model_sampling, steps)
    elif scheduler_name == "simple":
        sigmas = simple_scheduler(model_sampling, steps)
    elif scheduler_name == "ddim_uniform":
        sigmas = ddim_scheduler(model_sampling, steps)
    elif scheduler_name == "sgm_uniform":
        sigmas = normal_scheduler(model_sampling, steps, sgm=True)
    elif scheduler_name == "beta":
        sigmas = beta_scheduler(model_sampling, steps)
    else:
        logging.error("error invalid scheduler {}".format(scheduler_name))
    return sigmas

def sampler_object(name):
    if name == "uni_pc":
        sampler = KSAMPLER(uni_pc.sample_unipc)
    elif name == "uni_pc_bh2":
        sampler = KSAMPLER(uni_pc.sample_unipc_bh2)
    elif name == "ddim":
        sampler = ksampler("euler", inpaint_options={"random": True})
    else:
        sampler = ksampler(name)
    return sampler

class KSampler:
    SCHEDULERS = SCHEDULER_NAMES
    SAMPLERS = SAMPLER_NAMES
    DISCARD_PENULTIMATE_SIGMA_SAMPLERS = set(('dpm_2', 'dpm_2_ancestral', 'uni_pc', 'uni_pc_bh2'))

    def __init__(self, model, steps, device, sampler=None, scheduler=None, denoise=None, model_options={}):
        self.model = model
        self.device = device
        if scheduler not in self.SCHEDULERS:
            scheduler = self.SCHEDULERS[0]
        if sampler not in self.SAMPLERS:
            sampler = self.SAMPLERS[0]
        self.scheduler = scheduler
        self.sampler = sampler
        self.set_steps(steps, denoise)
        self.denoise = denoise
        self.model_options = model_options

    def calculate_sigmas(self, steps):
        sigmas = None

        discard_penultimate_sigma = False
        if self.sampler in self.DISCARD_PENULTIMATE_SIGMA_SAMPLERS:
            steps += 1
            discard_penultimate_sigma = True

        sigmas = calculate_sigmas(self.model.get_model_object("model_sampling"), self.scheduler, steps)

        if discard_penultimate_sigma:
            sigmas = torch.cat([sigmas[:-2], sigmas[-1:]])
        return sigmas

    def set_steps(self, steps, denoise=None):
        self.steps = steps
        if denoise is None or denoise > 0.9999:
            self.sigmas = self.calculate_sigmas(steps).to(self.device)
        else:
            if denoise <= 0.0:
                self.sigmas = torch.FloatTensor([])
            else:
                new_steps = int(steps/denoise)
                sigmas = self.calculate_sigmas(new_steps).to(self.device)
                self.sigmas = sigmas[-(steps + 1):]

    def sample(self, noise, positive, negative, cfg, latent_image=None, start_step=None, last_step=None, force_full_denoise=False, denoise_mask=None, sigmas=None, callback=None, disable_pbar=False, seed=None):
        if sigmas is None:
            sigmas = self.sigmas

        if last_step is not None and last_step < (len(sigmas) - 1):
            sigmas = sigmas[:last_step + 1]
            if force_full_denoise:
                sigmas[-1] = 0

        if start_step is not None:
            if start_step < (len(sigmas) - 1):
                sigmas = sigmas[start_step:]
            else:
                if latent_image is not None:
                    return latent_image
                else:
                    return torch.zeros_like(noise)

        sampler = sampler_object(self.sampler)

        return sample(self.model, noise, positive, negative, cfg, self.device, sampler, sigmas, self.model_options, latent_image=latent_image, denoise_mask=denoise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed)