Reduce the usage of GPU
#20
by
Fabrice-TIERCELIN
- opened
app.py
CHANGED
|
@@ -1,6 +1,8 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import torch
|
| 3 |
import os
|
|
|
|
|
|
|
| 4 |
from glob import glob
|
| 5 |
from pathlib import Path
|
| 6 |
from typing import Optional
|
|
@@ -9,9 +11,6 @@ from diffusers import StableVideoDiffusionPipeline
|
|
| 9 |
from diffusers.utils import export_to_video
|
| 10 |
from PIL import Image
|
| 11 |
|
| 12 |
-
import random
|
| 13 |
-
import spaces
|
| 14 |
-
|
| 15 |
fps25Pipe = StableVideoDiffusionPipeline.from_pretrained(
|
| 16 |
"vdo/stable-video-diffusion-img2vid-xt-1-1", torch_dtype=torch.float16, variant="fp16"
|
| 17 |
)
|
|
@@ -24,8 +23,7 @@ fps14Pipe.to("cuda")
|
|
| 24 |
|
| 25 |
max_64_bit_int = 2**63 - 1
|
| 26 |
|
| 27 |
-
|
| 28 |
-
def sample(
|
| 29 |
image: Image,
|
| 30 |
seed: Optional[int] = 42,
|
| 31 |
randomize_seed: bool = True,
|
|
@@ -35,7 +33,6 @@ def sample(
|
|
| 35 |
decoding_t: int = 3,
|
| 36 |
frame_format: str = "webp",
|
| 37 |
version: str = "auto",
|
| 38 |
-
device: str = "cuda",
|
| 39 |
output_folder: str = "outputs",
|
| 40 |
):
|
| 41 |
if image.mode == "RGBA":
|
|
@@ -43,20 +40,47 @@ def sample(
|
|
| 43 |
|
| 44 |
if randomize_seed:
|
| 45 |
seed = random.randint(0, max_64_bit_int)
|
| 46 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
|
| 48 |
os.makedirs(output_folder, exist_ok=True)
|
| 49 |
base_count = len(glob(os.path.join(output_folder, "*.mp4")))
|
| 50 |
video_path = os.path.join(output_folder, f"{base_count:06d}.mp4")
|
| 51 |
|
| 52 |
-
if version == "svdxt" or (14 < fps_id and version != "svd"):
|
| 53 |
-
frames = fps25Pipe(image, decode_chunk_size=decoding_t, generator=generator, motion_bucket_id=motion_bucket_id, noise_aug_strength=noise_aug_strength, num_frames=25).frames[0]
|
| 54 |
-
else:
|
| 55 |
-
frames = fps14Pipe(image, decode_chunk_size=decoding_t, generator=generator, motion_bucket_id=motion_bucket_id, noise_aug_strength=noise_aug_strength, num_frames=25).frames[0]
|
| 56 |
export_to_video(frames, video_path, fps=fps_id)
|
| 57 |
|
| 58 |
return video_path, gr.update(value=video_path, visible=True), gr.update(label="Generated frames in *." + frame_format + " format", format = frame_format, value = frames, visible=True), seed
|
| 59 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
def resize_image(image, output_size=(1024, 576)):
|
| 61 |
# Calculate aspect ratios
|
| 62 |
target_aspect = output_size[0] / output_size[1] # Aspect ratio of the desired size
|
|
@@ -117,7 +141,7 @@ with gr.Blocks() as demo:
|
|
| 117 |
gallery = gr.Gallery(label="Generated frames", visible=False)
|
| 118 |
|
| 119 |
image.upload(fn=resize_image, inputs=image, outputs=image, queue=False)
|
| 120 |
-
generate_btn.click(fn=
|
| 121 |
|
| 122 |
gr.Examples(
|
| 123 |
examples=[
|
|
@@ -127,7 +151,7 @@ with gr.Blocks() as demo:
|
|
| 127 |
],
|
| 128 |
inputs=[image, seed, randomize_seed, motion_bucket_id, fps_id, noise_aug_strength, decoding_t, frame_format, version],
|
| 129 |
outputs=[video, download_button, gallery, seed],
|
| 130 |
-
fn=
|
| 131 |
run_on_click=True,
|
| 132 |
cache_examples=False,
|
| 133 |
)
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import torch
|
| 3 |
import os
|
| 4 |
+
import random
|
| 5 |
+
import spaces
|
| 6 |
from glob import glob
|
| 7 |
from pathlib import Path
|
| 8 |
from typing import Optional
|
|
|
|
| 11 |
from diffusers.utils import export_to_video
|
| 12 |
from PIL import Image
|
| 13 |
|
|
|
|
|
|
|
|
|
|
| 14 |
fps25Pipe = StableVideoDiffusionPipeline.from_pretrained(
|
| 15 |
"vdo/stable-video-diffusion-img2vid-xt-1-1", torch_dtype=torch.float16, variant="fp16"
|
| 16 |
)
|
|
|
|
| 23 |
|
| 24 |
max_64_bit_int = 2**63 - 1
|
| 25 |
|
| 26 |
+
def animate(
|
|
|
|
| 27 |
image: Image,
|
| 28 |
seed: Optional[int] = 42,
|
| 29 |
randomize_seed: bool = True,
|
|
|
|
| 33 |
decoding_t: int = 3,
|
| 34 |
frame_format: str = "webp",
|
| 35 |
version: str = "auto",
|
|
|
|
| 36 |
output_folder: str = "outputs",
|
| 37 |
):
|
| 38 |
if image.mode == "RGBA":
|
|
|
|
| 40 |
|
| 41 |
if randomize_seed:
|
| 42 |
seed = random.randint(0, max_64_bit_int)
|
| 43 |
+
|
| 44 |
+
frames = animate_on_gpu(
|
| 45 |
+
image,
|
| 46 |
+
seed,
|
| 47 |
+
randomize_seed,
|
| 48 |
+
motion_bucket_id,
|
| 49 |
+
fps_id,
|
| 50 |
+
noise_aug_strength,
|
| 51 |
+
decoding_t,
|
| 52 |
+
frame_format,
|
| 53 |
+
version
|
| 54 |
+
)
|
| 55 |
|
| 56 |
os.makedirs(output_folder, exist_ok=True)
|
| 57 |
base_count = len(glob(os.path.join(output_folder, "*.mp4")))
|
| 58 |
video_path = os.path.join(output_folder, f"{base_count:06d}.mp4")
|
| 59 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
export_to_video(frames, video_path, fps=fps_id)
|
| 61 |
|
| 62 |
return video_path, gr.update(value=video_path, visible=True), gr.update(label="Generated frames in *." + frame_format + " format", format = frame_format, value = frames, visible=True), seed
|
| 63 |
|
| 64 |
+
@spaces.GPU(duration=120)
|
| 65 |
+
def animate_on_gpu(
|
| 66 |
+
image: Image,
|
| 67 |
+
seed: Optional[int] = 42,
|
| 68 |
+
randomize_seed: bool = True,
|
| 69 |
+
motion_bucket_id: int = 127,
|
| 70 |
+
fps_id: int = 6,
|
| 71 |
+
noise_aug_strength: float = 0.1,
|
| 72 |
+
decoding_t: int = 3,
|
| 73 |
+
frame_format: str = "webp",
|
| 74 |
+
version: str = "auto"
|
| 75 |
+
):
|
| 76 |
+
generator = torch.manual_seed(seed)
|
| 77 |
+
|
| 78 |
+
if version == "svdxt" or (14 < fps_id and version != "svd"):
|
| 79 |
+
return fps25Pipe(image, decode_chunk_size=decoding_t, generator=generator, motion_bucket_id=motion_bucket_id, noise_aug_strength=noise_aug_strength, num_frames=25).frames[0]
|
| 80 |
+
else:
|
| 81 |
+
return fps14Pipe(image, decode_chunk_size=decoding_t, generator=generator, motion_bucket_id=motion_bucket_id, noise_aug_strength=noise_aug_strength, num_frames=25).frames[0]
|
| 82 |
+
|
| 83 |
+
|
| 84 |
def resize_image(image, output_size=(1024, 576)):
|
| 85 |
# Calculate aspect ratios
|
| 86 |
target_aspect = output_size[0] / output_size[1] # Aspect ratio of the desired size
|
|
|
|
| 141 |
gallery = gr.Gallery(label="Generated frames", visible=False)
|
| 142 |
|
| 143 |
image.upload(fn=resize_image, inputs=image, outputs=image, queue=False)
|
| 144 |
+
generate_btn.click(fn=animate, inputs=[image, seed, randomize_seed, motion_bucket_id, fps_id, noise_aug_strength, decoding_t, frame_format, version], outputs=[video, download_button, gallery, seed], api_name="video")
|
| 145 |
|
| 146 |
gr.Examples(
|
| 147 |
examples=[
|
|
|
|
| 151 |
],
|
| 152 |
inputs=[image, seed, randomize_seed, motion_bucket_id, fps_id, noise_aug_strength, decoding_t, frame_format, version],
|
| 153 |
outputs=[video, download_button, gallery, seed],
|
| 154 |
+
fn=animate,
|
| 155 |
run_on_click=True,
|
| 156 |
cache_examples=False,
|
| 157 |
)
|