hpoghos commited on
Commit
da5ac73
·
verified ·
1 Parent(s): e882c67

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +6 -13
app.py CHANGED
@@ -6,7 +6,7 @@ from os.path import join as opj
6
  import argparse
7
  import datetime
8
  from pathlib import Path
9
- # import spaces
10
  import gradio as gr
11
  import tempfile
12
  import yaml
@@ -29,7 +29,6 @@ args = parser.parse_args()
29
  Path(args.where_to_log).mkdir(parents=True, exist_ok=True)
30
  result_fol = Path(args.where_to_log).absolute()
31
  device = args.device
32
- device_cpu = "cpu"
33
 
34
 
35
  # --------------------------
@@ -41,10 +40,10 @@ cfg_v2v = {'downscale': 1, 'upscale_size': (1280, 720), 'model_id': 'damo/Video-
41
  # --------------------------
42
  # ----- Initialization -----
43
  # --------------------------
44
- ms_model = init_modelscope(device_cpu)
45
  # # zs_model = init_zeroscope(device)
46
- ad_model = init_animatediff(device_cpu)
47
- svd_model = init_svd(device_cpu)
48
  sdxl_model = init_sdxl(device)
49
 
50
  ckpt_file_streaming_t2v = Path("t2v_enhanced/checkpoints/streaming_t2v.ckpt").absolute()
@@ -57,7 +56,7 @@ msxl_model = init_v2v_model(cfg_v2v)
57
  # -------------------------
58
  # ----- Functionality -----
59
  # -------------------------
60
- # @spaces.GPU(duration=120)
61
  def generate(prompt, num_frames, image, model_name_stage1, model_name_stage2, seed, t, image_guidance, where_to_log=result_fol):
62
  now = datetime.datetime.now()
63
  name = prompt[:100].replace(" ", "_") + "_" + str(now.time()).replace(":", "_").replace(".", "_")
@@ -74,26 +73,20 @@ def generate(prompt, num_frames, image, model_name_stage1, model_name_stage2, se
74
  inference_generator = torch.Generator(device="cuda").manual_seed(seed)
75
 
76
  if model_name_stage1 == "ModelScopeT2V (text to video)":
77
- ms_model.to(device)
78
  short_video = ms_short_gen(prompt, ms_model, inference_generator, t, device)
79
- ms_model.to(device_cpu)
80
  elif model_name_stage1 == "AnimateDiff (text to video)":
81
- ad_model.to(device)
82
  short_video = ad_short_gen(prompt, ad_model, inference_generator, t, device)
83
- ad_model.to(device_cpu)
84
  elif model_name_stage1 == "SVD (image to video)":
85
  # For cached examples
86
  if isinstance(image, dict):
87
  image = image["path"]
88
- svd_model.to(device)
89
  short_video = svd_short_gen(image, prompt, svd_model, sdxl_model, inference_generator, t, device)
90
- svd_model.to(device_cpu)
91
 
92
  stream_long_gen(prompt, short_video, n_autoreg_gen, seed, t, image_guidance, name, stream_cli, stream_model)
93
  video_path = opj(where_to_log, name+".mp4")
94
  return video_path
95
 
96
- # @spaces.GPU(duration=300)
97
  def enhance(prompt, input_to_enhance, num_frames=None, image=None, model_name_stage1=None, model_name_stage2=None, seed=33, t=50, image_guidance=9.5, result_fol=result_fol):
98
  if input_to_enhance is None:
99
  input_to_enhance = generate(prompt, num_frames, image, model_name_stage1, model_name_stage2, seed, t, image_guidance)
 
6
  import argparse
7
  import datetime
8
  from pathlib import Path
9
+ import spaces
10
  import gradio as gr
11
  import tempfile
12
  import yaml
 
29
  Path(args.where_to_log).mkdir(parents=True, exist_ok=True)
30
  result_fol = Path(args.where_to_log).absolute()
31
  device = args.device
 
32
 
33
 
34
  # --------------------------
 
40
  # --------------------------
41
  # ----- Initialization -----
42
  # --------------------------
43
+ ms_model = init_modelscope(device)
44
  # # zs_model = init_zeroscope(device)
45
+ ad_model = init_animatediff(device)
46
+ svd_model = init_svd(device)
47
  sdxl_model = init_sdxl(device)
48
 
49
  ckpt_file_streaming_t2v = Path("t2v_enhanced/checkpoints/streaming_t2v.ckpt").absolute()
 
56
  # -------------------------
57
  # ----- Functionality -----
58
  # -------------------------
59
+ @spaces.GPU(duration=120)
60
  def generate(prompt, num_frames, image, model_name_stage1, model_name_stage2, seed, t, image_guidance, where_to_log=result_fol):
61
  now = datetime.datetime.now()
62
  name = prompt[:100].replace(" ", "_") + "_" + str(now.time()).replace(":", "_").replace(".", "_")
 
73
  inference_generator = torch.Generator(device="cuda").manual_seed(seed)
74
 
75
  if model_name_stage1 == "ModelScopeT2V (text to video)":
 
76
  short_video = ms_short_gen(prompt, ms_model, inference_generator, t, device)
 
77
  elif model_name_stage1 == "AnimateDiff (text to video)":
 
78
  short_video = ad_short_gen(prompt, ad_model, inference_generator, t, device)
 
79
  elif model_name_stage1 == "SVD (image to video)":
80
  # For cached examples
81
  if isinstance(image, dict):
82
  image = image["path"]
 
83
  short_video = svd_short_gen(image, prompt, svd_model, sdxl_model, inference_generator, t, device)
 
84
 
85
  stream_long_gen(prompt, short_video, n_autoreg_gen, seed, t, image_guidance, name, stream_cli, stream_model)
86
  video_path = opj(where_to_log, name+".mp4")
87
  return video_path
88
 
89
+ @spaces.GPU(duration=400)
90
  def enhance(prompt, input_to_enhance, num_frames=None, image=None, model_name_stage1=None, model_name_stage2=None, seed=33, t=50, image_guidance=9.5, result_fol=result_fol):
91
  if input_to_enhance is None:
92
  input_to_enhance = generate(prompt, num_frames, image, model_name_stage1, model_name_stage2, seed, t, image_guidance)