File size: 4,613 Bytes
c60e9e5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
import spaces
import torch
import gradio as gr
import numpy as np
import random
from PIL import Image
from diffusers import CogVideoXImageToVideoPipeline
from diffusers.utils import export_to_video
import tempfile
import os

# Model configuration
MODEL_ID = "THUDM/CogVideoX-5b-I2V"
MAX_SEED = np.iinfo(np.int32).max

# Load pipeline globally (on CPU first, moved to GPU when needed)
print("Loading CogVideoX pipeline...")
pipe = CogVideoXImageToVideoPipeline.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.bfloat16,
)
pipe.enable_model_cpu_offload()
pipe.vae.enable_slicing()
pipe.vae.enable_tiling()
print("Pipeline loaded!")

def resize_image(image: Image.Image, max_size: int = 720) -> Image.Image:
    """Resize image to fit within max_size while maintaining aspect ratio."""
    width, height = image.size
    if max(width, height) > max_size:
        if width > height:
            new_width = max_size
            new_height = int(height * max_size / width)
        else:
            new_height = max_size
            new_width = int(width * max_size / height)
        # Make dimensions divisible by 16
        new_width = (new_width // 16) * 16
        new_height = (new_height // 16) * 16
        image = image.resize((new_width, new_height), Image.LANCZOS)
    return image

@spaces.GPU(duration=300)
def generate_video(
    image: Image.Image,
    prompt: str,
    negative_prompt: str = "",
    num_frames: int = 49,
    guidance_scale: float = 6.0,
    num_inference_steps: int = 50,
    seed: int = -1,
):
    """Generate video from image and prompt."""
    if image is None:
        raise gr.Error("Please upload an image!")

    if not prompt:
        prompt = "Make this image come alive with smooth, cinematic motion"

    # Set seed
    if seed == -1:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator(device="cuda").manual_seed(seed)

    # Resize image
    image = resize_image(image)

    # Move to GPU and generate
    pipe.to("cuda")

    with torch.inference_mode():
        video_frames = pipe(
            image=image,
            prompt=prompt,
            negative_prompt=negative_prompt,
            num_frames=num_frames,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            generator=generator,
        ).frames[0]

    # Export to video file
    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as f:
        export_to_video(video_frames, f.name, fps=8)
        return f.name, seed

# Gradio UI
with gr.Blocks(title="Video Generator") as demo:
    gr.Markdown("""
    # 🎬 Image to Video Generator

    Upload an image and describe the motion you want. Powered by CogVideoX.

    **Tips:**
    - Use clear, descriptive prompts about motion (e.g., "the person waves hello", "the flower blooms")
    - Keep images simple with clear subjects for best results
    """)

    with gr.Row():
        with gr.Column():
            image_input = gr.Image(type="pil", label="Upload Image")
            prompt_input = gr.Textbox(
                label="Prompt",
                placeholder="Describe the motion you want...",
                value="Make this image come alive with smooth, cinematic motion"
            )
            negative_prompt = gr.Textbox(
                label="Negative Prompt (optional)",
                placeholder="What to avoid...",
                value="blurry, low quality, distorted"
            )

            with gr.Row():
                num_frames = gr.Slider(
                    minimum=17, maximum=81, value=49, step=8,
                    label="Number of Frames"
                )
                guidance_scale = gr.Slider(
                    minimum=1.0, maximum=15.0, value=6.0, step=0.5,
                    label="Guidance Scale"
                )

            with gr.Row():
                num_steps = gr.Slider(
                    minimum=20, maximum=100, value=50, step=5,
                    label="Inference Steps"
                )
                seed_input = gr.Number(
                    value=-1, label="Seed (-1 for random)"
                )

            generate_btn = gr.Button("🎬 Generate Video", variant="primary")

        with gr.Column():
            video_output = gr.Video(label="Generated Video")
            seed_output = gr.Number(label="Seed Used")

    generate_btn.click(
        fn=generate_video,
        inputs=[image_input, prompt_input, negative_prompt, num_frames, guidance_scale, num_steps, seed_input],
        outputs=[video_output, seed_output]
    )

if __name__ == "__main__":
    demo.launch()