-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathcomposition_gen.py
166 lines (148 loc) · 6.11 KB
/
composition_gen.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
import argparse
from pathlib import Path
import torch
import json
from cache_functions import *
from MyCodes.MyFluxCompositionPipeline import FluxCompositionPipeline
from transformers import T5EncoderModel
from diffusers.utils import load_image
from MyCodes import MyFluxForward
import os
import types
from MyCodes.myutils import seed_everything
def get_next_number(dirname):
if not os.path.exists(dirname):
os.makedirs(dirname)
files = [f for f in os.listdir(dirname)]
if not files:
return 1
nums = [int(f.split('.')[0].split('-')[-1]) for f in files if f.split('.')[0].split('-')[-1].isdigit()]
return max(nums) + 1 if nums else 1
def parse_args():
parser = argparse.ArgumentParser(description='code for composition')
parser.add_argument('--weights_dir', type=str, default='/root/your-path/weights',
help='model weights directory')
parser.add_argument('--config_path', type=str,
default='configs/composition/example_config.json',
help='path of config file')
parser.add_argument('--img_config', type=str,
default='configs/composition/example_imgs.json',
help='path of image config file')
parser.add_argument('--output_dir', type=str,
default='test_outputs/composition',
help='output directory')
parser.add_argument('--use_predefine', type=bool,
default=False,
help='whether to use predefine')
return parser.parse_args()
def load_models(args, dtype=torch.bfloat16):
if args.use_predefine:
from MyCodes.FluxTransformer2DModel_PREDEFINE import FluxTransformer2DModel
else:
from MyCodes.FluxTransformer2DModel import FluxTransformer2DModel
transformer = FluxTransformer2DModel.from_single_file(
pretrained_model_link_or_path_or_dict=f"{args.weights_dir}/flux1-dev.safetensors",
config=f"{args.weights_dir}/transformer_config.json",
torch_dtype=dtype,
local_files_only=True)
text_encoder_2 = T5EncoderModel.from_pretrained(
args.weights_dir,
subfolder="text_encoder_2",
torch_dtype=dtype)
pipe = FluxCompositionPipeline.from_pretrained(
args.weights_dir,
transformer=None,
text_encoder_2=None,
torch_dtype=dtype)
pipe.transformer = transformer
pipe.text_encoder_2 = text_encoder_2
# bind forward function
pipe.transformer.forward = types.MethodType(MyFluxForward.forward, pipe.transformer)
pipe.to('cuda')
return pipe
def generate_image(pipe, img_config, param_config, output_dir):
main_image = load_image(img_config["main_image"])
ref_image = load_image(img_config["ref_image"])
ref_segment = load_image(img_config["ref_segment"])
height=512
width=512
for param in param_config['params']:
if 'cache_type' in param:
ratio_scheduler = 'constant'
use_attn_map=False
if param['cache_type'] == 'ours_cache':
cache_type = 'ours_cache'
elif param['cache_type'] == 'ours_predefine':
cache_type = 'ours_predefine'
model_kwargs = {
'fresh_ratio': param['fresh_ratio'],
'cache_type': cache_type,
'ratio_scheduler': ratio_scheduler,
'force_fresh': 'global',
'fresh_threshold': param['fresh_threshold'],
'soft_fresh_weight': param['soft_fresh_weight'],
'tailing_step': param['tailing_step'],
'edit_base':2,
'hw': (height//16,width//16)
}
edit_idx = None if param['cascade_num']==0 else edit_region_parser(
img_config['x1'], img_config['y1'],
img_config['x2'], img_config['y2'],
cascade_num=param['cascade_num'],
height=height,
width=width)
cache_dic, current = cache_init(
model_kwargs,
param['num_inference_steps'],
edit_idx)
current['edit_idx_merged']=convert_to_cache_index(edit_idx,edit_base=2,bonus_ratio=0.8,height=height,width=width)
current['edit_idx_merged']=current['edit_idx_merged'].to("cuda")
if cache_type=='ours_predefine':
predefine_cache_fresh_indices(cache_dic, current)
joint_attention_kwargs = {
'use_attn_map': use_attn_map,
'cache_dic': cache_dic,
'use_cache': param['use_cache'],
'current': current,
}
torch.manual_seed(42)
res = pipe.gen(
prompt=img_config["prompt"],
main_image=main_image,
ref_image=ref_image,
ref_segment=ref_segment,
height=512,
width=512,
x1=img_config["x1"], y1=img_config["y1"],
x2=img_config["x2"], y2=img_config["y2"],
num_inference_steps=param['num_inference_steps'],
joint_attention_kwargs=joint_attention_kwargs,
use_rf_inversion=param['use_rf_inversion'],
eta=param['eta'],
gamma=param['gamma'],
start_timestep=param['start_timestep'],
stop_timestep=param['stop_timestep'],
blend_ratio=param['blend_ratio'],
generator=torch.Generator(device='cuda').manual_seed(42),
skip_T=3 if 'inv_skip' not in param else param['inv_skip']
)
image=res.images[0]
num = get_next_number(output_dir)
image.save(f"{output_dir}/{num:03d}.png")
def main():
args = parse_args()
# ensure output directory exists
Path(args.output_dir).mkdir(parents=True,exist_ok=True)
# load model
pipe= load_models(args)
# load config file
with open(args.img_config, 'r') as f:
img_configs = json.load(f)
with open(args.config_path, 'r') as f:
param_config = json.load(f)
seed_everything()
# process each image
for img_config in img_configs['imgs']:
generate_image(pipe, img_config, param_config, args.output_dir)
if __name__ == "__main__":
main()