File size: 5,837 Bytes
7d5de5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8a98406
7d5de5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ad9f6fa
 
7d5de5f
47a907f
 
 
 
 
7d5de5f
47a907f
7d5de5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ad9f6fa
dd2d342
d840c30
 
ad9f6fa
7d5de5f
e70dd9f
7d5de5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
20e7c98
d840c30
7d5de5f
 
808dc9c
 
 
 
 
 
 
 
 
 
 
8a98406
808dc9c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8a98406
808dc9c
 
 
 
 
 
abd6d57
808dc9c
 
 
 
 
 
 
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
import torch
from diffusers.pipelines import FluxPipeline
from OminiControl.src.flux.condition import Condition
from PIL import Image
import random

from OminiControl.src.flux.generate import generate, seed_everything

print("Loading model...")
pipe = FluxPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16
)
pipe = pipe.to("cuda")

pipe.unload_lora_weights()

pipe.load_lora_weights(
    "vzhizhi6611/OminiControlArt",
    weight_name=f"v0/ghibli.safetensors",
    adapter_name="ghibli",
)
pipe.load_lora_weights(
    "vzhizhi6611/OminiControlArt",
    weight_name=f"v0/irasutoya.safetensors",
    adapter_name="irasutoya",
)
pipe.load_lora_weights(
    "vzhizhi6611/OminiControlArt",
    weight_name=f"v0/simpsons.safetensors",
    adapter_name="simpsons",
)
pipe.load_lora_weights(
    "vzhizhi6611/OminiControlArt",
    weight_name=f"v0/snoopy.safetensors",
    adapter_name="snoopy",
)

def generate_image(
    image,
    style,
    inference_mode,
    image_guidance,
    image_ratio,
    steps,
    use_random_seed,
    seed,
):
    pipe.enable_lora()
    # Prepare Condition
    def resize(img, factor=16):
        w, h = img.size
        new_w, new_h = w // factor * factor, h // factor * factor
        padding_w, padding_h = (w - new_w) // 2, (h - new_h) // 2
        img = img.crop((padding_w, padding_h, new_w + padding_w, new_h + padding_h))
        return img

    # Set Adapter
    activate_adapter_name = {
        "Studio Ghibli": "ghibli",
        "Irasutoya Illustration": "irasutoya",
        "The Simpsons": "simpsons",
        "Snoopy": "snoopy",
    }[style]
    pipe.set_adapters(activate_adapter_name)

    original_width, original_height = image.size

    factor = 512 / max(image.size)
    image = resize(
        image.resize(
            (int(image.size[0] * factor), int(image.size[1] * factor)),
            Image.LANCZOS,
        )
    )

    delta = -image.size[0] // 16
    condition = Condition(
        "subject",
        # activate_adapter_name,
        image,
        position_delta=(0, delta),
    )

    # Prepare seed
    if use_random_seed:
        seed = random.randint(0, 2**32 - 1)
    seed_everything(seed)

    # Image guidance scale
    image_guidance = 1.0 if inference_mode == "Fast" else image_guidance

    # Output size
    if image_ratio == "Auto":
        r = image.size[0] / image.size[1]
        ratio = min([0.67, 1, 1.5], key=lambda x: abs(x - r))
    else:
        ratio = {
            "Square(1:1)": 1,
            "Portrait(2:3)": 0.67,
            "Landscape(3:2)": 1.5,
        }[image_ratio]
    width, height = {
        0.67: (640, 960),
        1: (640, 640),
        1.5: (960, 640),
    }[ratio]


    output_factor = max(width, height) / max(original_width, original_height)
    width = int(original_width * output_factor)
    height = int(original_height * output_factor)

    print(
        f"Image Ratio: {image_ratio}, Inference Mode: {inference_mode}, Image Guidance: {image_guidance}, Seed: {seed}, Steps: {steps}, Ratio: {ratio}, Size: {width}x{height}"
    )
    # Generate
    result_img = generate(
        pipe,
        prompt="",
        conditions=[condition],
        num_inference_steps=steps,
        width=width,
        height=height,
        image_guidance_scale=image_guidance,
        default_lora=True,
        max_sequence_length=32,
    ).images[0]
    # result_img = image

    result_img = result_img.resize((width, height), Image.LANCZOS)

    return result_img


def generate_image_with_prompt(
    image,
    prompt,
    inference_mode,
    image_guidance,
    image_ratio,
    steps,
    use_random_seed,
    seed,
):
    pipe.disable_lora()
    # Prepare Condition
    def resize(img, factor=16):
        w, h = img.size
        new_w, new_h = w // factor * factor, h // factor * factor
        padding_w, padding_h = (w - new_w) // 2, (h - new_h) // 2
        img = img.crop((padding_w, padding_h, new_w + padding_w, new_h + padding_h))
        return img

    original_width, original_height = image.size

    factor = 512 / max(image.size)
    image = resize(
        image.resize(
            (int(image.size[0] * factor), int(image.size[1] * factor)),
            Image.LANCZOS,
        )
    )

    delta = -image.size[0] // 16
    condition = Condition(
        "subject",
        # activate_adapter_name,
        image,
        position_delta=(0, delta),
    )

    # Prepare seed
    if use_random_seed:
        seed = random.randint(0, 2**32 - 1)
    seed_everything(seed)

    # Image guidance scale
    image_guidance = 1.0 if inference_mode == "Fast" else image_guidance

    # Output size
    if image_ratio == "Auto":
        r = image.size[0] / image.size[1]
        ratio = min([0.67, 1, 1.5], key=lambda x: abs(x - r))
    else:
        ratio = {
            "Square(1:1)": 1,
            "Portrait(2:3)": 0.67,
            "Landscape(3:2)": 1.5,
        }[image_ratio]
    width, height = {
        0.67: (640, 960),
        1: (640, 640),
        1.5: (960, 640),
    }[ratio]


    output_factor = max(width, height) / max(original_width, original_height)
    width = int(original_width * output_factor)
    height = int(original_height * output_factor)

    print(
        f"Image Ratio: {image_ratio}, Inference Mode: {inference_mode}, Image Guidance: {image_guidance}, Seed: {seed}, Steps: {steps}, Ratio: {ratio}, Size: {width}x{height}"
    )
    # Generate
    result_img = generate(
        pipe,
        prompt=prompt,
        conditions=[condition],
        num_inference_steps=steps,
        width=width,
        height=height,
        image_guidance_scale=image_guidance,
        default_lora=True,
        max_sequence_length=32,
    ).images[0]
    # result_img = image

    result_img = result_img.resize((width, height), Image.LANCZOS)

    return result_img