Spaces:
Running
on
Zero
Running
on
Zero
Commit
•
ad0cea8
1
Parent(s):
52b0321
Update app.py
Browse files
app.py
CHANGED
@@ -8,6 +8,7 @@ from pyramid_dit import PyramidDiTForVideoGeneration
|
|
8 |
from diffusers.utils import export_to_video
|
9 |
|
10 |
import spaces
|
|
|
11 |
|
12 |
import subprocess
|
13 |
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
|
@@ -62,49 +63,62 @@ model = load_model()
|
|
62 |
|
63 |
# Text-to-video generation function
|
64 |
@spaces.GPU(duration=240)
|
65 |
-
def generate_video(prompt, duration, guidance_scale, video_guidance_scale):
|
66 |
temp = int(duration * 2.4) # Convert seconds to temp value (assuming 24 FPS)
|
67 |
torch_dtype = torch.bfloat16 if MODEL_DTYPE == "bf16" else torch.float32
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
84 |
export_to_video(frames, output_path, fps=24)
|
85 |
return output_path
|
86 |
|
87 |
# Image-to-video generation function
|
88 |
-
|
89 |
-
def generate_video_from_image(image, prompt, duration, video_guidance_scale):
|
90 |
-
temp = int(duration * 2.4) # Convert seconds to temp value (assuming 24 FPS)
|
91 |
-
torch_dtype = torch.bfloat16 if MODEL_DTYPE == "bf16" else torch.float32
|
92 |
-
|
93 |
-
target_size = (1280, 720)
|
94 |
-
cropped_image = center_crop(image, 1280, 720)
|
95 |
-
resized_image = cropped_image.resize((1280, 720))
|
96 |
-
|
97 |
-
with torch.no_grad(), torch.cuda.amp.autocast(enabled=True, dtype=torch_dtype):
|
98 |
-
frames = model.generate_i2v(
|
99 |
-
prompt=prompt,
|
100 |
-
input_image=resized_image,
|
101 |
-
num_inference_steps=[10, 10, 10],
|
102 |
-
temp=temp,
|
103 |
-
guidance_scale=7.0,
|
104 |
-
video_guidance_scale=video_guidance_scale,
|
105 |
-
output_type="pil",
|
106 |
-
save_memory=True,
|
107 |
-
)
|
108 |
|
109 |
output_path = "output_video_i2v.mp4"
|
110 |
export_to_video(frames, output_path, fps=24)
|
@@ -114,38 +128,41 @@ def generate_video_from_image(image, prompt, duration, video_guidance_scale):
|
|
114 |
with gr.Blocks() as demo:
|
115 |
gr.Markdown("# Pyramid Flow Video Generation Demo")
|
116 |
|
117 |
-
with gr.Tab("Text-to-Video"):
|
118 |
-
|
119 |
-
|
120 |
-
|
|
|
|
|
|
|
121 |
t2v_duration = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="Duration (seconds)")
|
122 |
t2v_guidance_scale = gr.Slider(minimum=1, maximum=15, value=9, step=0.1, label="Guidance Scale")
|
123 |
t2v_video_guidance_scale = gr.Slider(minimum=1, maximum=15, value=5, step=0.1, label="Video Guidance Scale")
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
t2v_generate_btn.click(
|
129 |
-
generate_video,
|
130 |
-
inputs=[t2v_prompt, t2v_duration, t2v_guidance_scale, t2v_video_guidance_scale],
|
131 |
-
outputs=t2v_output
|
132 |
-
)
|
133 |
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
144 |
|
145 |
-
i2v_generate_btn.click(
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
)
|
150 |
|
151 |
demo.launch()
|
|
|
8 |
from diffusers.utils import export_to_video
|
9 |
|
10 |
import spaces
|
11 |
+
import uuid
|
12 |
|
13 |
import subprocess
|
14 |
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
|
|
|
63 |
|
64 |
# Text-to-video generation function
|
65 |
@spaces.GPU(duration=240)
|
66 |
+
def generate_video(image, prompt, duration, guidance_scale, video_guidance_scale):
|
67 |
temp = int(duration * 2.4) # Convert seconds to temp value (assuming 24 FPS)
|
68 |
torch_dtype = torch.bfloat16 if MODEL_DTYPE == "bf16" else torch.float32
|
69 |
+
if(image):
|
70 |
+
cropped_image = center_crop(image, 1280, 720)
|
71 |
+
resized_image = cropped_image.resize((1280, 720))
|
72 |
+
with torch.no_grad(), torch.cuda.amp.autocast(enabled=True, dtype=torch_dtype):
|
73 |
+
frames = model.generate_i2v(
|
74 |
+
prompt=prompt,
|
75 |
+
input_image=resized_image,
|
76 |
+
num_inference_steps=[10, 10, 10],
|
77 |
+
temp=temp,
|
78 |
+
guidance_scale=7.0,
|
79 |
+
video_guidance_scale=video_guidance_scale,
|
80 |
+
output_type="pil",
|
81 |
+
save_memory=True,
|
82 |
+
)
|
83 |
+
else:
|
84 |
+
with torch.no_grad(), torch.cuda.amp.autocast(enabled=True, dtype=torch_dtype):
|
85 |
+
frames = model.generate(
|
86 |
+
prompt=prompt,
|
87 |
+
num_inference_steps=[20, 20, 20],
|
88 |
+
video_num_inference_steps=[10, 10, 10],
|
89 |
+
height=768,
|
90 |
+
width=1280,
|
91 |
+
temp=temp,
|
92 |
+
guidance_scale=guidance_scale,
|
93 |
+
video_guidance_scale=video_guidance_scale,
|
94 |
+
output_type="pil",
|
95 |
+
save_memory=True,
|
96 |
+
)
|
97 |
+
output_path = f"{str(uuid.uuid4())}_output_video.mp4"
|
98 |
export_to_video(frames, output_path, fps=24)
|
99 |
return output_path
|
100 |
|
101 |
# Image-to-video generation function
|
102 |
+
#@spaces.GPU(duration=240)
|
103 |
+
#def generate_video_from_image(image, prompt, duration, video_guidance_scale):
|
104 |
+
# temp = int(duration * 2.4) # Convert seconds to temp value (assuming 24 FPS)
|
105 |
+
# torch_dtype = torch.bfloat16 if MODEL_DTYPE == "bf16" else torch.float32
|
106 |
+
#
|
107 |
+
# target_size = (1280, 720)
|
108 |
+
# cropped_image = center_crop(image, 1280, 720)
|
109 |
+
# resized_image = cropped_image.resize((1280, 720))
|
110 |
+
#
|
111 |
+
# with torch.no_grad(), torch.cuda.amp.autocast(enabled=True, dtype=torch_dtype):
|
112 |
+
# frames = model.generate_i2v(
|
113 |
+
# prompt=prompt,
|
114 |
+
# input_image=resized_image,
|
115 |
+
# num_inference_steps=[10, 10, 10],
|
116 |
+
# temp=temp,
|
117 |
+
# guidance_scale=7.0,
|
118 |
+
# video_guidance_scale=video_guidance_scale,
|
119 |
+
# output_type="pil",
|
120 |
+
# save_memory=True,
|
121 |
+
# )
|
122 |
|
123 |
output_path = "output_video_i2v.mp4"
|
124 |
export_to_video(frames, output_path, fps=24)
|
|
|
128 |
with gr.Blocks() as demo:
|
129 |
gr.Markdown("# Pyramid Flow Video Generation Demo")
|
130 |
|
131 |
+
#with gr.Tab("Text-to-Video"):
|
132 |
+
with gr.Row():
|
133 |
+
with gr.Column():
|
134 |
+
with gr.Accordion("Image to Video (optional)", open=False):
|
135 |
+
i2v_image = gr.Image(type="pil", label="Input Image")
|
136 |
+
t2v_prompt = gr.Textbox(label="Prompt")
|
137 |
+
with gr.Accordion("Advanced settings", open=False):
|
138 |
t2v_duration = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="Duration (seconds)")
|
139 |
t2v_guidance_scale = gr.Slider(minimum=1, maximum=15, value=9, step=0.1, label="Guidance Scale")
|
140 |
t2v_video_guidance_scale = gr.Slider(minimum=1, maximum=15, value=5, step=0.1, label="Video Guidance Scale")
|
141 |
+
t2v_generate_btn = gr.Button("Generate Video")
|
142 |
+
with gr.Column():
|
143 |
+
t2v_output = gr.Video(label="Generated Video")
|
|
|
|
|
|
|
|
|
|
|
|
|
144 |
|
145 |
+
t2v_generate_btn.click(
|
146 |
+
generate_video,
|
147 |
+
inputs=[i2v_image, t2v_prompt, t2v_duration, t2v_guidance_scale, t2v_video_guidance_scale],
|
148 |
+
outputs=t2v_output
|
149 |
+
)
|
150 |
+
|
151 |
+
#with gr.Tab("Image-to-Video"):
|
152 |
+
# with gr.Row():
|
153 |
+
# with gr.Column():
|
154 |
+
|
155 |
+
# i2v_prompt = gr.Textbox(label="Prompt")
|
156 |
+
# i2v_duration = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="Duration (seconds)")
|
157 |
+
# i2v_video_guidance_scale = gr.Slider(minimum=1, maximum=15, value=4, step=0.1, label="Video Guidance Scale")
|
158 |
+
# i2v_generate_btn = gr.Button("Generate Video")
|
159 |
+
# with gr.Column():
|
160 |
+
# i2v_output = gr.Video(label="Generated Video")
|
161 |
|
162 |
+
#i2v_generate_btn.click(
|
163 |
+
# generate_video_from_image,
|
164 |
+
# inputs=[i2v_image, i2v_prompt, i2v_duration, i2v_video_guidance_scale],
|
165 |
+
# outputs=i2v_output
|
166 |
+
#)
|
167 |
|
168 |
demo.launch()
|