-
Notifications
You must be signed in to change notification settings - Fork 119
/
inference_magictime.py
194 lines (166 loc) · 7.95 KB
/
inference_magictime.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
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
import os
import json
import torch
import inspect
import argparse
import pandas as pd
from omegaconf import OmegaConf
from transformers import CLIPTextModel, CLIPTokenizer
from diffusers import AutoencoderKL, DDIMScheduler
from diffusers.utils.import_utils import is_xformers_available
from utils.unet import UNet3DConditionModel
from utils.pipeline_magictime import MagicTimePipeline
from utils.util import save_videos_grid
from utils.util import load_weights
@torch.no_grad()
def main(args):
*_, func_args = inspect.getargvalues(inspect.currentframe())
func_args = dict(func_args)
if 'counter' not in globals():
globals()['counter'] = 0
unique_id = globals()['counter']
globals()['counter'] += 1
savedir = None
savedir = os.path.join(args.save_path, f"{unique_id}")
while os.path.exists(savedir):
unique_id = globals()['counter']
globals()['counter'] += 1
savedir = os.path.join(args.save_path, f"{unique_id}")
os.makedirs(savedir, exist_ok=True)
print(f"The results will be save to {savedir}")
model_config = OmegaConf.load(args.config)[0]
inference_config = OmegaConf.load(args.config)[1]
if model_config.magic_adapter_s_path:
print("Use MagicAdapter-S")
if model_config.magic_adapter_t_path:
print("Use MagicAdapter-T")
if model_config.magic_text_encoder_path:
print("Use Magic_Text_Encoder")
tokenizer = CLIPTokenizer.from_pretrained(model_config.pretrained_model_path, subfolder="tokenizer")
text_encoder = CLIPTextModel.from_pretrained(model_config.pretrained_model_path, subfolder="text_encoder").cuda()
vae = AutoencoderKL.from_pretrained(model_config.pretrained_model_path, subfolder="vae").cuda()
unet = UNet3DConditionModel.from_pretrained_2d(model_config.pretrained_model_path, subfolder="unet",
unet_additional_kwargs=OmegaConf.to_container(
inference_config.unet_additional_kwargs)).cuda()
if is_xformers_available() and (not args.without_xformers):
unet.enable_xformers_memory_efficient_attention()
pipeline = MagicTimePipeline(
vae=vae, text_encoder=text_encoder, tokenizer=tokenizer, unet=unet,
scheduler=DDIMScheduler(**OmegaConf.to_container(inference_config.noise_scheduler_kwargs)),
).to("cuda")
pipeline = load_weights(
pipeline,
motion_module_path=model_config.get("motion_module", ""),
dreambooth_model_path=model_config.get("dreambooth_path", ""),
magic_adapter_s_path=model_config.get("magic_adapter_s_path", ""),
magic_adapter_t_path=model_config.get("magic_adapter_t_path", ""),
magic_text_encoder_path=model_config.get("magic_text_encoder_path", ""),
).to("cuda")
if args.human:
sample_idx = 0
while True:
user_prompt = input("Enter your prompt (or type 'exit' to quit): ")
if user_prompt.lower() == "exit":
break
random_seed = torch.randint(0, 2 ** 32 - 1, (1,)).item()
torch.manual_seed(random_seed)
print(f"current seed: {random_seed}")
print(f"sampling {user_prompt} ...")
sample = pipeline(
user_prompt,
negative_prompt = list(model_config.n_prompt),
num_inference_steps = model_config.steps,
guidance_scale = model_config.guidance_scale,
width = model_config.W,
height = model_config.H,
video_length = model_config.L,
).videos
prompt_for_filename = "-".join(user_prompt.replace("/", "").split(" ")[:10])
save_videos_grid(sample, f"{savedir}/sample/{sample_idx}-{random_seed}-{prompt_for_filename}.mp4")
print(f"save to {savedir}/sample/{sample_idx}-{random_seed}-{prompt_for_filename}.mp4")
sample_idx += 1
else:
default = True
batch_size = args.batch_size
if args.run_csv:
print("run csv")
default = False
file_path = args.run_csv
data = pd.read_csv(file_path)
prompts = data['name'].tolist()
videoids = data['videoid'].tolist()
elif args.run_json:
print("run json")
default = False
file_path = args.run_json
with open(file_path, 'r') as file:
data = json.load(file)
prompts = []
videoids = []
senids = []
for item in data['sentences']:
prompts.append(item['caption'])
videoids.append(item['video_id'])
senids.append(item['sen_id'])
elif args.run_txt:
print("run txt")
default = False
file_path = args.run_txt
with open(file_path, 'r') as file:
prompts = [line.strip() for line in file.readlines()]
videoids = [f"video_{i}" for i in range(len(prompts))]
else:
prompts = model_config.prompt
videoids = [f"video_{i}" for i in range(len(prompts))]
for i in range(0, len(prompts), batch_size):
batch_prompts_raw = prompts[i : i + batch_size]
batch_prompts = [prompt for prompt in batch_prompts_raw]
if args.run_csv or args.run_json or args.run_txt or default:
batch_videoids = videoids[i : i + batch_size]
if args.run_json:
batch_senids = senids[i : i + batch_size]
flag = True
for idx in range(len(batch_prompts)):
if args.run_csv or args.run_txt or default:
new_filename = f"{batch_videoids[idx]}.mp4"
if args.run_json:
new_filename = f"{batch_videoids[idx]}-{batch_senids[idx]}.mp4"
if not os.path.exists(os.path.join(savedir, new_filename)):
flag = False
break
if flag:
print("skipping")
continue
n_prompts = list(model_config.n_prompt) * len(batch_prompts) if len(model_config.n_prompt) == 1 else model_config.n_prompt
random_seed = torch.randint(0, 2**32 - 1, (1,)).item()
torch.manual_seed(random_seed)
print(f"current seed: {random_seed}")
results = pipeline(
batch_prompts,
negative_prompt = n_prompts,
num_inference_steps = model_config.steps,
guidance_scale = model_config.guidance_scale,
width = model_config.W,
height = model_config.H,
video_length = model_config.L,
).videos
for idx, sample in enumerate(results):
if args.run_csv or args.run_txt or default:
new_filename = f"{batch_videoids[idx]}.mp4"
if args.run_json:
new_filename = f"{batch_videoids[idx]}-{batch_senids[idx]}.mp4"
save_videos_grid(sample.unsqueeze(0), f"{savedir}/{new_filename}")
print(f"save to {savedir}/{new_filename}")
OmegaConf.save(model_config, f"{savedir}/model_config.yaml")
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--config", type=str, required=True)
parser.add_argument("--without-xformers", action="store_true")
parser.add_argument("--human", action="store_true", help="Enable human mode for interactive video generation")
parser.add_argument("--run-csv", type=str, default=None)
parser.add_argument("--run-json", type=str, default=None)
parser.add_argument("--run-txt", type=str, default=None)
parser.add_argument("--save-path", type=str, default="outputs")
parser.add_argument("--batch-size", type=int, default=1)
args = parser.parse_args()
main(args)