Spaces:
Running
on
Zero
Running
on
Zero
removed orientation mode
Browse files- gradio/app.py +6 -34
gradio/app.py
CHANGED
|
@@ -11,7 +11,7 @@ from diffusers.utils import export_to_video
|
|
| 11 |
from inference import load_model, inference_on_image
|
| 12 |
|
| 13 |
# -----------------------
|
| 14 |
-
# 1. Load
|
| 15 |
# -----------------------
|
| 16 |
args = argparse.Namespace()
|
| 17 |
args.blur2vid_hf_repo_path = "tedlasai/blur2vid"
|
|
@@ -19,8 +19,6 @@ args.pretrained_model_path = "THUDM/CogVideoX-2b"
|
|
| 19 |
args.model_config_path = "training/configs/outsidephotos.yaml"
|
| 20 |
args.video_width = 1280
|
| 21 |
args.video_height = 720
|
| 22 |
-
# args.video_width = 960
|
| 23 |
-
# args.video_height = 540
|
| 24 |
args.seed = None
|
| 25 |
|
| 26 |
pipe, model_config = load_model(args)
|
|
@@ -29,8 +27,8 @@ OUTPUT_DIR = Path("/tmp/generated_videos")
|
|
| 29 |
OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
|
| 30 |
|
| 31 |
|
| 32 |
-
@spaces.GPU(timeout=300)
|
| 33 |
-
def generate_video_from_image(image: Image.Image, interval_key: str,
|
| 34 |
"""
|
| 35 |
Wrapper for Gradio. Takes an image and returns a video path.
|
| 36 |
"""
|
|
@@ -43,19 +41,6 @@ def generate_video_from_image(image: Image.Image, interval_key: str, orientation
|
|
| 43 |
print("Device:", torch.cuda.get_device_name(0))
|
| 44 |
print("bf16 supported:", torch.cuda.is_bf16_supported())
|
| 45 |
|
| 46 |
-
if orientation_mode == "Landscape (1280Γ720)":
|
| 47 |
-
print("Chosing resolution 1280Γ720 (landscape)")
|
| 48 |
-
args.video_width = 1280
|
| 49 |
-
args.video_height = 720
|
| 50 |
-
elif orientation_mode == "Portrait (720Γ1280)":
|
| 51 |
-
print("Choosing resolution 720Γ1280 (portrait)")
|
| 52 |
-
args.video_height = 1280
|
| 53 |
-
args.video_width = 720
|
| 54 |
-
else:
|
| 55 |
-
print("Unknown orientation mode", orientation_mode, "defaulting to 1280x720")
|
| 56 |
-
args.video_width = 1280
|
| 57 |
-
args.video_height = 720
|
| 58 |
-
|
| 59 |
args.num_inference_steps = num_inference_steps
|
| 60 |
|
| 61 |
video_id = uuid.uuid4().hex
|
|
@@ -84,7 +69,8 @@ with gr.Blocks(css="footer {visibility: hidden}") as demo:
|
|
| 84 |
- π **Project page:** <https://blur2vid.github.io/>
|
| 85 |
- π» **Code:** <https://github.com/tedlasai/blur2vid/>
|
| 86 |
|
| 87 |
-
Upload a blurry image and the model will generate a short video
|
|
|
|
| 88 |
"""
|
| 89 |
)
|
| 90 |
|
|
@@ -104,20 +90,6 @@ with gr.Blocks(css="footer {visibility: hidden}") as demo:
|
|
| 104 |
interactive=True,
|
| 105 |
)
|
| 106 |
|
| 107 |
-
with gr.Row():
|
| 108 |
-
mode_choice = gr.Radio(
|
| 109 |
-
label="Orientation",
|
| 110 |
-
choices=["Landscape (1280Γ720)", "Portrait (720Γ1280)"],
|
| 111 |
-
value="Landscape (1280Γ720)",
|
| 112 |
-
interactive=True,
|
| 113 |
-
)
|
| 114 |
-
|
| 115 |
-
gr.Markdown(
|
| 116 |
-
"<span style='font-size: 12px; color: gray;'>"
|
| 117 |
-
"Note: Model was trained on 1280Γ720 (Landscape). Portrait mode will degrade performance."
|
| 118 |
-
"</span>"
|
| 119 |
-
)
|
| 120 |
-
|
| 121 |
num_inference_steps = gr.Slider(
|
| 122 |
label="Number of inference steps",
|
| 123 |
minimum=4,
|
|
@@ -139,7 +111,7 @@ with gr.Blocks(css="footer {visibility: hidden}") as demo:
|
|
| 139 |
|
| 140 |
generate_btn.click(
|
| 141 |
fn=generate_video_from_image,
|
| 142 |
-
inputs=[image_in, tense_choice,
|
| 143 |
outputs=video_out,
|
| 144 |
api_name="predict",
|
| 145 |
)
|
|
|
|
| 11 |
from inference import load_model, inference_on_image
|
| 12 |
|
| 13 |
# -----------------------
|
| 14 |
+
# 1. Load model
|
| 15 |
# -----------------------
|
| 16 |
args = argparse.Namespace()
|
| 17 |
args.blur2vid_hf_repo_path = "tedlasai/blur2vid"
|
|
|
|
| 19 |
args.model_config_path = "training/configs/outsidephotos.yaml"
|
| 20 |
args.video_width = 1280
|
| 21 |
args.video_height = 720
|
|
|
|
|
|
|
| 22 |
args.seed = None
|
| 23 |
|
| 24 |
pipe, model_config = load_model(args)
|
|
|
|
| 27 |
OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
|
| 28 |
|
| 29 |
|
| 30 |
+
@spaces.GPU(timeout=300, duration=200)
|
| 31 |
+
def generate_video_from_image(image: Image.Image, interval_key: str, num_inference_steps: int) -> str:
|
| 32 |
"""
|
| 33 |
Wrapper for Gradio. Takes an image and returns a video path.
|
| 34 |
"""
|
|
|
|
| 41 |
print("Device:", torch.cuda.get_device_name(0))
|
| 42 |
print("bf16 supported:", torch.cuda.is_bf16_supported())
|
| 43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
args.num_inference_steps = num_inference_steps
|
| 45 |
|
| 46 |
video_id = uuid.uuid4().hex
|
|
|
|
| 69 |
- π **Project page:** <https://blur2vid.github.io/>
|
| 70 |
- π» **Code:** <https://github.com/tedlasai/blur2vid/>
|
| 71 |
|
| 72 |
+
Upload a blurry image and the model will generate a short video showing the recovered motion based on your selection.
|
| 73 |
+
Note: The image will be resized to 1280Γ720. We recommend uploading landscape-oriented images.
|
| 74 |
"""
|
| 75 |
)
|
| 76 |
|
|
|
|
| 90 |
interactive=True,
|
| 91 |
)
|
| 92 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
num_inference_steps = gr.Slider(
|
| 94 |
label="Number of inference steps",
|
| 95 |
minimum=4,
|
|
|
|
| 111 |
|
| 112 |
generate_btn.click(
|
| 113 |
fn=generate_video_from_image,
|
| 114 |
+
inputs=[image_in, tense_choice, num_inference_steps],
|
| 115 |
outputs=video_out,
|
| 116 |
api_name="predict",
|
| 117 |
)
|