app.py

import spaces
import gradio as gr
import torch
import os
from glob import glob
from pathlib import Path
from typing import Optional
from huggingface_hub import HfFolder
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 hf_hub_download

title = '''# 👋🏻Welcome to Tonic's🌟🎥StableVideo XT-1-1 
🌟🎥StableVideo XT-1-1 (SVD) Image-to-Video is a latent diffusion model trained to generate short video clips from an image conditioning. Check out the [Community demo for Stable Video Diffusion](https://huggingface.co/spaces/multimodalart/stable-video-diffusion) - Img2Vid - XT ([model](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt-1-1), [paper](https://stability.ai/research/stable-video-diffusion-scaling-latent-video-diffusion-models-to-large-datasets), [stability's ui waitlist](https://stability.ai/contact))
#### Research release ([_non-commercial_](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt/blob/main/LICENSE)): generate `4s` vid from a single image at (`25 frames` at `6 fps`). this demo uses [🧨 diffusers for low VRAM and fast generation](https://huggingface.co/docs/diffusers/main/en/using-diffusers/svd).
Join us : 🌟TeamTonic🌟 is always making cool demos! Join our active builder's🛠️community 👻  [![Join us on Discord](https://img.shields.io/discord/1109943800132010065?label=Discord&logo=discord&style=flat-square)](https://discord.gg/GWpVpekp) On 🤗Huggingface: [TeamTonic](https://huggingface.co/TeamTonic) & [MultiTransformer](https://huggingface.co/MultiTransformer) On 🌐Github: [Tonic-AI](https://github.com/tonic-ai) & contribute to 🌟 [SciTonic](https://github.com/Tonic-AI/scitonic) 🤗Big thanks to Yuvi Sharma and all the folks at huggingface for the community grant 🤗
'''

# Load the API token from an environment variable
hf_token = os.getenv("HF_TOKEN")

# If the token is not found, raise an error or handle it appropriately
if not hf_token:
    raise ValueError("Hugging Face token not found. Please set the HF_TOKEN environment variable.")

# Use the token for authentication
HfFolder.save_token(hf_token)

# Load the original model to cache
original_model_id = "stabilityai/stable-video-diffusion-img2vid-xt"
pipe = StableVideoDiffusionPipeline.from_pretrained(
    original_model_id,
    torch_dtype=torch.float16,
    variant="fp16",
    use_auth_token=hf_token
)

# Get the cache directory of the original model
cache_dir = HfFolder.get_cache_dir()
model_cache_dir = os.path.join(cache_dir, original_model_id.replace("/", "--"))

# Path to the downloaded svd_xt_1_1.safetensors file
downloaded_safetensors_path = "svd_xt_1_1.safetensors"

# Replace the original safetensors file with the downloaded one
original_safetensors_path = os.path.join(model_cache_dir, "svd_xt.safetensors")
os.replace(downloaded_safetensors_path, original_safetensors_path)

# Load the model from the local directory with the replaced safetensors file
pipe = StableVideoDiffusionPipeline.from_pretrained(
    model_cache_dir,
    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 sample(
    image: Image,
    seed: Optional[int] = 42,
    randomize_seed: bool = True,
    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 image.mode == "RGBA":
        image = image.convert("RGB")
        
    if(randomize_seed):
        seed = random.randint(0, max_64_bit_int)
    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=14).frames[0]
    export_to_video(frames, video_path, fps=fps_id)
    torch.manual_seed(seed)
    
    return video_path, seed

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:
  gr.Markdown(title)
  with gr.Row():
    with gr.Column():
        image = gr.Image(label="Upload your image", type="pil")
        generate_btn = gr.Button("Generate")
    video = gr.Video()
  with gr.Accordion("Advanced options", open=False):
      seed = gr.Slider(label="Seed", value=42, randomize=True, minimum=0, maximum=max_64_bit_int, step=1)
      randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
      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)
      
  image.upload(fn=resize_image, inputs=image, outputs=image, queue=False)
  generate_btn.click(fn=sample, inputs=[image, seed, randomize_seed, motion_bucket_id, fps_id], outputs=[video, seed], api_name="video")
  gr.Examples(
    examples=[
        "images/blink_meme.png",
        "images/confused2_meme.png",
        "images/disaster_meme.png",
        "images/distracted_meme.png",
        "images/hide_meme.png",
        "images/nazare_meme.png",
        "images/success_meme.png",
        "images/willy_meme.png",
        "images/wink_meme.png"
    ],
    inputs=image,
    outputs=[video, seed],
    fn=sample,
    cache_examples=True,
  )

if __name__ == "__main__":
    demo.queue(max_size=20)
    demo.launch(share=True)