app.py

import subprocess
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
# subprocess.run('pip install bitsandbytes', shell=True)
subprocess.run('pip install av==12.0.0', shell=True)

import gradio as gr
import spaces
#import gradio.helpers
import torch
import os
from glob import glob
from pathlib import Path
from typing import Optional

# from diffusers import StableVideoDiffusionPipeline
from kandinsky import get_T2V_pipeline
from diffusers.utils import load_image, export_to_video
from PIL import Image

import uuid
import random
from huggingface_hub import hf_hub_download

#gradio.helpers.CACHED_FOLDER = '/data/cache'

# pipe = StableVideoDiffusionPipeline.from_pretrained(
#     "multimodalart/stable-video-diffusion", 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)

device_map = {
    "dit": torch.device('cuda'), 
    "vae": torch.device('cuda'), 
    "text_embedder": torch.device('cuda')
}
pipe = get_T2V_pipeline(device_map)

max_64_bit_int = 2**63 - 1

@spaces.GPU(duration=120)
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",
    progress=gr.Progress(track_tqdm=True)
):
    # 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=25).frames[0]
    prompt = "The camera follows behind a white vintage SUV with a black roof rack as it speeds up a steep dirt road surrounded by pine trees on a steep mountain slope, dust kicks up from it’s tires, the sunlight shines on the SUV as it speeds along the dirt road, casting a warm glow over the scene. The dirt road curves gently into the distance, with no other cars or vehicles in sight. The trees on either side of the road are redwoods, with patches of greenery scattered throughout. The car is seen from the rear following the curve with ease, making it seem as if it is on a rugged drive through the rugged terrain. The dirt road itself is surrounded by steep hills and mountains, with a clear blue sky above with wispy clouds."
    frames = pipe(
        seed=seed,
        time_length=12,
        width = 672,
        height = 384,
        save_path=video_path,
        text=prompt,
    )
    # export_to_video(frames, video_path, fps=8)
    torch.manual_seed(seed)

    return video_path

def resize_image(image, output_size=(672, 384)):
    # 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('''# Community demo for Stable Video Diffusion - Img2Vid - XT ([model](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt), [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).
  ''')
  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=[], outputs=[video], 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="lazy",
  # )

if __name__ == "__main__":
    #demo.queue(max_size=20, api_open=False)
    demo.launch(share=True, show_api=False)