rahul7star commited on
Commit
aeb9017
·
verified ·
1 Parent(s): 07fdcce

new math testing

Browse files
Files changed (1) hide show
  1. app.py +56 -13
app.py CHANGED
@@ -34,24 +34,66 @@ pipe.load_lora_weights(causvid_path, adapter_name="causvid_lora")
34
  pipe.set_adapters(["causvid_lora"], adapter_weights=[0.95])
35
  pipe.fuse_lora()
36
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37
  MOD_VALUE = 32
38
- DEFAULT_H_SLIDER_VALUE = 512
39
- DEFAULT_W_SLIDER_VALUE = 896
40
 
41
- # Environment variable check
42
- IS_ORIGINAL_SPACE = os.environ.get("IS_ORIGINAL_SPACE", "True") == "True"
 
43
 
44
- # Original limits
45
- ORIGINAL_SLIDER_MIN_H, ORIGINAL_SLIDER_MAX_H = 128, 1280
46
- ORIGINAL_SLIDER_MIN_W, ORIGINAL_SLIDER_MAX_W = 128, 1280
47
- ORIGINAL_MAX_DURATION = round(81/24, 1) # MAX_FRAMES_MODEL/FIXED_FPS
48
 
49
- # Limited space constants
50
  LIMITED_MAX_RESOLUTION = 640
51
  LIMITED_MAX_DURATION = 2.0
52
  LIMITED_MAX_STEPS = 4
53
 
54
- # Set limits based on environment variable
 
 
 
 
 
 
55
  if IS_ORIGINAL_SPACE:
56
  SLIDER_MIN_H, SLIDER_MAX_H = 128, LIMITED_MAX_RESOLUTION
57
  SLIDER_MIN_W, SLIDER_MAX_W = 128, LIMITED_MAX_RESOLUTION
@@ -61,14 +103,15 @@ else:
61
  SLIDER_MIN_H, SLIDER_MAX_H = ORIGINAL_SLIDER_MIN_H, ORIGINAL_SLIDER_MAX_H
62
  SLIDER_MIN_W, SLIDER_MAX_W = ORIGINAL_SLIDER_MIN_W, ORIGINAL_SLIDER_MAX_W
63
  MAX_DURATION = ORIGINAL_MAX_DURATION
64
- MAX_STEPS = 8
65
 
66
  MAX_SEED = np.iinfo(np.int32).max
67
 
68
  FIXED_FPS = 24
69
- FIXED_OUTPUT_FPS = 18 # we downspeed the output video as a temporary "trick"
70
  MIN_FRAMES_MODEL = 8
71
- MAX_FRAMES_MODEL = 81
 
72
 
73
  default_prompt_t2v = "cinematic footage, group of pedestrians dancing in the streets of NYC, high quality breakdance, 4K, tiktok video, intricate details, instagram feel, dynamic camera, smooth dance motion, dimly lit, stylish, beautiful faces, smiling, music video"
74
  default_negative_prompt = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards, watermark, text, signature"
 
34
  pipe.set_adapters(["causvid_lora"], adapter_weights=[0.95])
35
  pipe.fuse_lora()
36
 
37
+ # MOD_VALUE = 32
38
+ # DEFAULT_H_SLIDER_VALUE = 512
39
+ # DEFAULT_W_SLIDER_VALUE = 896
40
+
41
+ # # Environment variable check
42
+ # IS_ORIGINAL_SPACE = os.environ.get("IS_ORIGINAL_SPACE", "True") == "True"
43
+
44
+ # # Original limits
45
+ # ORIGINAL_SLIDER_MIN_H, ORIGINAL_SLIDER_MAX_H = 128, 1280
46
+ # ORIGINAL_SLIDER_MIN_W, ORIGINAL_SLIDER_MAX_W = 128, 1280
47
+ # ORIGINAL_MAX_DURATION = round(81/24, 1) # MAX_FRAMES_MODEL/FIXED_FPS
48
+
49
+ # # Limited space constants
50
+ # LIMITED_MAX_RESOLUTION = 640
51
+ # LIMITED_MAX_DURATION = 2.0
52
+ # LIMITED_MAX_STEPS = 4
53
+
54
+ # # Set limits based on environment variable
55
+ # if IS_ORIGINAL_SPACE:
56
+ # SLIDER_MIN_H, SLIDER_MAX_H = 128, LIMITED_MAX_RESOLUTION
57
+ # SLIDER_MIN_W, SLIDER_MAX_W = 128, LIMITED_MAX_RESOLUTION
58
+ # MAX_DURATION = LIMITED_MAX_DURATION
59
+ # MAX_STEPS = LIMITED_MAX_STEPS
60
+ # else:
61
+ # SLIDER_MIN_H, SLIDER_MAX_H = ORIGINAL_SLIDER_MIN_H, ORIGINAL_SLIDER_MAX_H
62
+ # SLIDER_MIN_W, SLIDER_MAX_W = ORIGINAL_SLIDER_MIN_W, ORIGINAL_SLIDER_MAX_W
63
+ # MAX_DURATION = ORIGINAL_MAX_DURATION
64
+ # MAX_STEPS = 8
65
+
66
+ # MAX_SEED = np.iinfo(np.int32).max
67
+
68
+ # FIXED_FPS = 24
69
+ # FIXED_OUTPUT_FPS = 18 # we downspeed the output video as a temporary "trick"
70
+ # MIN_FRAMES_MODEL = 8
71
+ # MAX_FRAMES_MODEL = 81
72
+
73
+
74
+ #New math to make it High Res
75
+
76
  MOD_VALUE = 32
 
 
77
 
78
+ # Defaults for higher-res generation
79
+ DEFAULT_H_SLIDER_VALUE = 768
80
+ DEFAULT_W_SLIDER_VALUE = 1344 # 16:9 friendly and divisible by MOD_VALUE
81
 
82
+ # Original Space = Hugging Face space with compute limits
83
+ IS_ORIGINAL_SPACE = os.environ.get("IS_ORIGINAL_SPACE", "True") == "True"
 
 
84
 
85
+ # Conservative limits for low-end environments
86
  LIMITED_MAX_RESOLUTION = 640
87
  LIMITED_MAX_DURATION = 2.0
88
  LIMITED_MAX_STEPS = 4
89
 
90
+ # Generous limits for local or Pro spaces
91
+ ORIGINAL_SLIDER_MIN_H, ORIGINAL_SLIDER_MAX_H = 128, 1536
92
+ ORIGINAL_SLIDER_MIN_W, ORIGINAL_SLIDER_MAX_W = 128, 1536
93
+ ORIGINAL_MAX_DURATION = round(81 / 24, 1) # 3.4 seconds
94
+ ORIGINAL_MAX_STEPS = 8
95
+
96
+ # Use limited or original (generous) settings
97
  if IS_ORIGINAL_SPACE:
98
  SLIDER_MIN_H, SLIDER_MAX_H = 128, LIMITED_MAX_RESOLUTION
99
  SLIDER_MIN_W, SLIDER_MAX_W = 128, LIMITED_MAX_RESOLUTION
 
103
  SLIDER_MIN_H, SLIDER_MAX_H = ORIGINAL_SLIDER_MIN_H, ORIGINAL_SLIDER_MAX_H
104
  SLIDER_MIN_W, SLIDER_MAX_W = ORIGINAL_SLIDER_MIN_W, ORIGINAL_SLIDER_MAX_W
105
  MAX_DURATION = ORIGINAL_MAX_DURATION
106
+ MAX_STEPS = ORIGINAL_MAX_STEPS
107
 
108
  MAX_SEED = np.iinfo(np.int32).max
109
 
110
  FIXED_FPS = 24
111
+ FIXED_OUTPUT_FPS = 18 # reduce final video FPS to save space
112
  MIN_FRAMES_MODEL = 8
113
+ MAX_FRAMES_MODEL = 81
114
+
115
 
116
  default_prompt_t2v = "cinematic footage, group of pedestrians dancing in the streets of NYC, high quality breakdance, 4K, tiktok video, intricate details, instagram feel, dynamic camera, smooth dance motion, dimly lit, stylish, beautiful faces, smiling, music video"
117
  default_negative_prompt = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards, watermark, text, signature"