-
Notifications
You must be signed in to change notification settings - Fork 5
/
ipadapter_sdxl_plus.py
172 lines (152 loc) · 5.57 KB
/
ipadapter_sdxl_plus.py
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
from diffusers import StableDiffusionXLPipeline, DDIMScheduler
import torch
from PIL import Image
import config as cfg
from ip_adapter.ip_adapter import IPAdapter
device = "cuda"
pipe = StableDiffusionXLPipeline.from_single_file(cfg.sdxl_base_model_path, torch_dtype=torch.float16)
pipe.watermark = None
pipe.safety_checker = None
pipe.feature_extractor = None
pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
pipe.to(device)
image1 = Image.open("assets/input_venere.jpg")
image2 = Image.open("assets/input_portrait.jpg")
image3 = Image.open("assets/input_warrior.jpg")
ip_adapter = IPAdapter(pipe, cfg.ipadapter_sdxl_plus_vit_h_path, cfg.image_encoder_sd15_path, device=device)
generator = torch.Generator().manual_seed(1)
"""
SDXL Plus model with one reference image and text prompt
"""
prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds = ip_adapter.get_prompt_embeds(
image1,
prompt="beautiful renaissance woman",
negative_prompt="blurry, horror, worst quality, low quality",
)
image = pipe(
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
num_inference_steps=30,
guidance_scale=5.0,
num_images_per_prompt=1,
height=1024,
width=1024,
generator=generator,
).images[0]
image.save("output/ipadapter_sdxl_plus.webp", lossless=True, quality=100)
"""
SDXL Plus model with three reference images and text prompt
"""
prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds = ip_adapter.get_prompt_embeds(
[image1, image2, image3],
prompt="beautiful renaissance warrior woman",
negative_prompt="blurry, horror, worst quality, low quality",
)
image = pipe(
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
num_inference_steps=30,
guidance_scale=5.0,
num_images_per_prompt=1,
height=1024,
width=1024,
generator=generator,
).images[0]
image.save("output/ipadapter_sdxl_plus_multi.webp", lossless=True, quality=100)
"""
SDXL plus model with three reference images and noisy negative images
Negative image can be anything but it seems to react better to very noisy images
"""
noise = Image.effect_noise((224, 224), 10)
prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds = ip_adapter.get_prompt_embeds(
[image1, image2, image3],
prompt="beautiful renaissance warrior woman",
negative_prompt="blurry, horror, worst quality, low quality",
negative_images=[noise, noise, noise],
)
image = pipe(
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
num_inference_steps=30,
guidance_scale=5.0,
num_images_per_prompt=1,
height=1024,
width=1024,
generator=generator,
).images[0]
image.save("output/ipadapter_sdxl_plus_noise.webp", lossless=True, quality=100)
"""
SDXL plus model with one reference image and text prompt
"""
prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds = ip_adapter.get_prompt_embeds(
image1,
prompt="beautiful woman wearing sunglasses",
negative_prompt="blurry, horror, worst quality, low quality",
)
ip_adapter.set_scale(.5)
image = pipe(
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
num_inference_steps=30,
guidance_scale=5.0,
num_images_per_prompt=1,
height=1024,
width=1024,
generator=generator,
).images[0]
image.save("output/ipadapter_sdxl_plus_text.webp", lossless=True, quality=100)
"""
SDXL Plus model with two weighted reference images and noise
"""
prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds = ip_adapter.get_prompt_embeds(
[image1, image2],
prompt="beautiful renaissance woman",
negative_prompt="blurry, horror, worst quality, low quality",
weight=[1.0, .9],
negative_images=[noise, noise],
)
ip_adapter.set_scale(1.0)
image = pipe(
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
num_inference_steps=30,
guidance_scale=4.0,
num_images_per_prompt=1,
height=1024,
width=1024,
generator=generator,
).images[0]
image.save("output/ipadapter_sdxl_plus_weight1.webp", lossless=True, quality=100)
"""
SDXL Plus model with two weighted reference images and noise
"""
prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds = ip_adapter.get_prompt_embeds(
[image1, image2],
prompt="beautiful renaissance woman",
negative_prompt="blurry, horror, worst quality, low quality",
weight=[.9, 1.0],
negative_images=[noise, noise],
)
image = pipe(
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
num_inference_steps=30,
guidance_scale=4.0,
num_images_per_prompt=1,
height=1024,
width=1024,
generator=generator,
).images[0]
image.save("output/ipadapter_sdxl_plus_weight2.webp", lossless=True, quality=100)