Create app.py

app.py

@@ -0,0 +1,143 @@
+import gradio as gr
+#import gradio.helpers
+import torch
+import os
+import base64
+from glob import glob
+from pathlib import Path
+from typing import Optional
+
+from diffusers import StableVideoDiffusionPipeline
+from diffusers.utils import load_image, export_to_video
+from PIL import Image
+
+import uuid
+import random
+from huggingface_hub import login, hf_hub_download
+import space
+
+#gradio.helpers.CACHED_FOLDER = '/data/cache'
+
+# SECRET_TOKEN = os.getenv('SECRET_TOKEN', 'default_secret')
+
+# HF_API_KEY = os.getenv('HF_API_KEY', '')
+# login(token=HF_API_KEY)
+
+pipe = StableVideoDiffusionPipeline.from_pretrained(
+    "stabilityai/stable-video-diffusion-img2vid-xt-1-1",
+    torch_dtype=torch.float16,
+    variant="fp16"
+)
+pipe.to("cuda")
+pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+#pipe.vae = torch.compile(pipe.vae, mode="reduce-overhead", fullgraph=True)
+
+max_64_bit_int = 2**63 - 1
+
+@spaces.GPU(enable_queue=True)
+def generate_video(
+    secret_token: str,
+    image: Image,
+    seed: int,
+    motion_bucket_id: int = 127,
+    fps_id: int = 6,
+    version: str = "svd_xt",
+    cond_aug: float = 0.02,
+    decoding_t: int = 3,  # Number of frames decoded at a time! This eats most VRAM. Reduce if necessary.
+    device: str = "cuda",
+    output_folder: str = "outputs",
+):
+    # if secret_token != SECRET_TOKEN:
+    #     raise gr.Error(
+    #         f'Invalid secret token. Please fork the original space if you want to use it for yourself.')
+
+
+    # note julian: normally we should resize input images, but normally they are already in 1024x576, so..
+    # also, I would like to experiment with vertical videos, and 1024x512 videos
+    image = resize_image(image)
+    
+    if image.mode == "RGBA":
+        image = image.convert("RGB")
+        
+    generator = torch.manual_seed(seed)
+    
+    os.makedirs(output_folder, exist_ok=True)
+    base_count = len(glob(os.path.join(output_folder, "*.mp4")))
+    video_path = os.path.join(output_folder, f"{base_count:06d}.mp4")
+
+    frames = pipe(image, decode_chunk_size=decoding_t, generator=generator, motion_bucket_id=motion_bucket_id, noise_aug_strength=0.1, num_frames=25).frames[0]
+    export_to_video(frames, video_path, fps=fps_id)
+    torch.manual_seed(seed)
+
+    # Read the content of the video file and encode it to base64
+    with open(video_path, "rb") as video_file:
+        video_base64 = base64.b64encode(video_file.read()).decode('utf-8')
+
+    # Prepend the appropriate data URI header with MIME type
+    video_data_uri = 'data:video/mp4;base64,' + video_base64
+    
+    # clean-up (otherwise there is a risk of "ghosting", eg. someone seeing the previous generated video",
+    # of one of the steps go wrong)
+    os.remove(video_path)
+    
+    return video_data_uri
+
+def resize_image(image, output_size=(1024, 576)):
+    # Calculate aspect ratios
+    target_aspect = output_size[0] / output_size[1]  # Aspect ratio of the desired size
+    image_aspect = image.width / image.height  # Aspect ratio of the original image
+
+    # Resize then crop if the original image is larger
+    if image_aspect > target_aspect:
+        # Resize the image to match the target height, maintaining aspect ratio
+        new_height = output_size[1]
+        new_width = int(new_height * image_aspect)
+        resized_image = image.resize((new_width, new_height), Image.LANCZOS)
+        # Calculate coordinates for cropping
+        left = (new_width - output_size[0]) / 2
+        top = 0
+        right = (new_width + output_size[0]) / 2
+        bottom = output_size[1]
+    else:
+        # Resize the image to match the target width, maintaining aspect ratio
+        new_width = output_size[0]
+        new_height = int(new_width / image_aspect)
+        resized_image = image.resize((new_width, new_height), Image.LANCZOS)
+        # Calculate coordinates for cropping
+        left = 0
+        top = (new_height - output_size[1]) / 2
+        right = output_size[0]
+        bottom = (new_height + output_size[1]) / 2
+
+    # Crop the image
+    cropped_image = resized_image.crop((left, top, right, bottom))
+    return cropped_image
+
+with gr.Blocks() as demo:
+    # secret_token = gr.Text(
+    #     label='Secret Token',
+    #     max_lines=1,
+    #     placeholder='Enter your secret token')
+    gr.HTML("""
+        <div style="z-index: 100; position: fixed; top: 0px; right: 0px; left: 0px; bottom: 0px; width: 100%; height: 100%; background: white; display: flex; align-items: center; justify-content: center; color: black;">
+        <div style="text-align: center; color: black;">
+        <p style="color: black;">This space is a REST API to programmatically generate MP4 videos.</p>
+        <p style="color: black;">Interested in using it? Look no further than the <a href="https://huggingface.co/spaces/multimodalart/stable-video-diffusion" target="_blank">original space</a>!</p>
+        </div>
+        </div>""")
+    image = gr.Image(label="Upload your image", type="pil")
+    generate_btn = gr.Button("Generate")
+    base64_out = gr.Textbox(label="Base64 Video")
+    seed = gr.Slider(label="Seed", value=42, randomize=False, minimum=0, maximum=max_64_bit_int, step=1)
+    motion_bucket_id = gr.Slider(label="Motion bucket id", info="Controls how much motion to add/remove from the image", value=127, minimum=1, maximum=255)
+    fps_id = gr.Slider(label="Frames per second", info="The length of your video in seconds will be 25/fps", value=6, minimum=5, maximum=30)
+       
+    generate_btn.click(
+        fn=generate_video,
+        # inputs=[secret_token, image, seed, motion_bucket_id, fps_id],
+        inputs=[image, seed, motion_bucket_id, fps_id],
+        outputs=base64_out,
+        api_name="run"
+    )
+
+demo.queue(max_size=20).launch()