File size: 3,931 Bytes
7d5de5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
325d5f4
462cace
51075f9
 
 
462cace
325d5f4
462cace
3e8686f
 
51075f9
462cace
7d5de5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
51075f9
 
7d5de5f
 
 
ad9f6fa
 
7d5de5f
47a907f
 
 
 
 
7d5de5f
47a907f
7d5de5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ad9f6fa
dd2d342
d840c30
 
ad9f6fa
7d5de5f
e70dd9f
7d5de5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
20e7c98
d840c30
7d5de5f
 
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
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",
)
# ref: https://civitai.com/models/715472/flux-hayao-miyazaki-ghibli
pipe.load_lora_weights(
    "./lora",
    weight_name="MaoMu_Ghibli.safetensors",
    adapter_name="MaoMu_Ghibli",
)
# ref: https://civitai.com/models/824739/flux-3d-animation-style-lora
pipe.load_lora_weights(
    "./lora",
    weight_name="3d_animation.safetensors",
    adapter_name="3d_animation",
)


def generate_image(
    image,
    style,
    inference_mode,
    image_guidance,
    image_ratio,
    steps,
    use_random_seed,
    seed,
):
    # 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",
        "3D Animation": "3d_animation",
        "MaoMu Ghibli": "MaoMu_Ghibli",
    }[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