File size: 4,707 Bytes
7bed60d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from modules.prompt_parser import SdConditioning

from base64 import b64decode as decode
from io import BytesIO as bIO
from PIL import Image
import numpy as np
import torch

from scripts.ui_funcs import parse_mapping


def empty_tensor(H: int, W: int):
    return torch.zeros((H, W)).unsqueeze(0)


def advanced_mapping(sd_model, couples: list, WIDTH: int, HEIGHT: int, mapping: list):
    data = parse_mapping(mapping)
    assert len(couples) == len(data)

    ARGs: dict = {}
    IS_SDXL: bool = hasattr(
        sd_model.forge_objects.unet.model.diffusion_model, "label_emb"
    )

    for tile_index in range(len(data)):
        mask = torch.zeros((HEIGHT, WIDTH))

        (X, Y, W) = data[tile_index]
        x_from = int(WIDTH * X[0])
        x_to = int(WIDTH * X[1])
        y_from = int(HEIGHT * Y[0])
        y_to = int(HEIGHT * Y[1])
        weight = W

        # ===== Cond =====
        texts = SdConditioning([couples[tile_index]], False, WIDTH, HEIGHT, None)
        cond = sd_model.get_learned_conditioning(texts)
        pos_cond = [[cond["crossattn"]]] if IS_SDXL else [[cond]]
        # ===== Cond =====

        # ===== Mask =====
        mask[y_from:y_to, x_from:x_to] = weight
        # ===== Mask =====

        ARGs[f"cond_{tile_index + 1}"] = pos_cond
        ARGs[f"mask_{tile_index + 1}"] = mask.unsqueeze(0)

    return ARGs


def basic_mapping(
    sd_model,
    couples: list,
    WIDTH: int,
    HEIGHT: int,
    LINE_COUNT: int,
    IS_HORIZONTAL: bool,
    background: str,
    TILE_SIZE: int,
    TILE_WEIGHT: float,
    BG_WEIGHT: float,
):

    ARGs: dict = {}
    IS_SDXL: bool = hasattr(
        sd_model.forge_objects.unet.model.diffusion_model, "label_emb"
    )

    for tile in range(LINE_COUNT):
        mask = torch.zeros((HEIGHT, WIDTH))

        # ===== Cond =====
        texts = SdConditioning([couples[tile]], False, WIDTH, HEIGHT, None)
        cond = sd_model.get_learned_conditioning(texts)
        pos_cond = [[cond["crossattn"]]] if IS_SDXL else [[cond]]
        # ===== Cond =====

        # ===== Mask =====
        if background == "First Line":
            if tile == 0:
                mask = torch.ones((HEIGHT, WIDTH)) * BG_WEIGHT
            else:
                if IS_HORIZONTAL:
                    mask[:, (tile - 1) * TILE_SIZE : tile * TILE_SIZE] = TILE_WEIGHT
                else:
                    mask[(tile - 1) * TILE_SIZE : tile * TILE_SIZE, :] = TILE_WEIGHT
        else:
            if IS_HORIZONTAL:
                mask[:, tile * TILE_SIZE : (tile + 1) * TILE_SIZE] = TILE_WEIGHT
            else:
                mask[tile * TILE_SIZE : (tile + 1) * TILE_SIZE, :] = TILE_WEIGHT
        # ===== Mask =====

        ARGs[f"cond_{tile + 1}"] = pos_cond
        ARGs[f"mask_{tile + 1}"] = mask.unsqueeze(0)

    if background == "Last Line":
        ARGs[f"mask_{LINE_COUNT}"] = (
            torch.ones((HEIGHT, WIDTH)) * BG_WEIGHT
        ).unsqueeze(0)

    return ARGs


def b64image2tensor(img: str, WIDTH: int, HEIGHT: int) -> torch.Tensor:
    image_bytes = decode(img)
    image = Image.open(bIO(image_bytes)).convert("L")

    if image.width != WIDTH or image.height != HEIGHT:
        image = image.resize((WIDTH, HEIGHT), resample=Image.Resampling.NEAREST)

    image = np.array(image).astype(np.float32) / 255.0
    image = torch.from_numpy(image).unsqueeze(0)

    return image


def mask_mapping(
    sd_model,
    couples: list,
    WIDTH: int,
    HEIGHT: int,
    LINE_COUNT: int,
    mapping: list[dict],
    background: str,
    BG_WEIGHT: float,
):

    mapping = [
        b64image2tensor(m["mask"], WIDTH, HEIGHT) * float(m["weight"]) for m in mapping
    ]

    ARGs: dict = {}
    IS_SDXL: bool = hasattr(
        sd_model.forge_objects.unet.model.diffusion_model, "label_emb"
    )

    for layer in range(LINE_COUNT):
        mask = torch.zeros((HEIGHT, WIDTH))

        # ===== Cond =====
        texts = SdConditioning([couples[layer]], False, WIDTH, HEIGHT, None)
        cond = sd_model.get_learned_conditioning(texts)
        pos_cond = [[cond["crossattn"]]] if IS_SDXL else [[cond]]
        # ===== Cond =====

        # ===== Mask =====
        if background == "First Line":
            if layer == 0:
                mask = torch.ones((HEIGHT, WIDTH)) * BG_WEIGHT
            else:
                mask = mapping[layer - 1]
        else:
            mask = mapping[layer]
        # ===== Mask =====

        ARGs[f"cond_{layer + 1}"] = pos_cond
        ARGs[f"mask_{layer + 1}"] = mask.unsqueeze(0) if mask.dim() == 2 else mask

    if background == "Last Line":
        ARGs[f"mask_{LINE_COUNT}"] = (
            torch.ones((HEIGHT, WIDTH)) * BG_WEIGHT
        ).unsqueeze(0)

    return ARGs