openvideo-ai 0.1.0

Open-source AI video generation library without API keys

OpenVideo

Open-source AI video generation library without API keys. Generate videos from text prompts or images using open-source models.

Features

• Text-to-Video: Generate videos from text descriptions
• Image-to-Video: Animate static images into videos
• Multiple Models: Support for various open-source models (Zeroscope, ModelScope, AnimateDiff)
• No API Keys Required: Fully local generation using open-source models
• CLI Interface: Easy command-line usage
• Python API: Programmatic access for developers

Installation

pip install openvideo

For development installation:

pip install openvideo[dev]

Requirements

• Python 3.9+
• CUDA-capable GPU (recommended for performance)
• 8GB+ VRAM for video generation

Quick Start

CLI Usage

# Text to Video
openvideo text2video -p "a cat walking in the snow" -o output.mp4

# Image to Video
openvideo image2video -i input.png -o output.mp4 -p "camera panning"

Python API

from openvideo import VideoGenerator

# Initialize generator
generator = VideoGenerator(model_name="zeroscope")

# Generate from text
frames = generator.generate(
    prompt="a beautiful sunset over the ocean",
    num_frames=24,
    width=512,
    height=512,
)

# Save video
generator.save_video(frames, "output.mp4", fps=24)

# Generate from image
from PIL import Image

img = Image.open("input.png")
frames = generator.generate_from_image(
    image=img,
    prompt="gentle camera pan",
    num_frames=24,
)
generator.save_video(frames, "output.mp4")

Supported Models

Model	Description	VRAM
zeroscope	High-quality text-to-video	8GB
modelscope	Fast text-to-video	6GB
animatediff	Animation from SD models	6GB

Configuration

GPU Settings

# Use specific device
generator = VideoGenerator(device="cuda")

# Use float16 for less VRAM
generator = VideoGenerator(dtype="float16")

Generation Parameters

Parameter	Default	Description
num_frames	24	Number of frames
width	512	Frame width
height	512	Frame height
num_inference_steps	25	Denoising steps
guidance_scale	7.5	CFG scale
seed	None	Random seed

Advanced Usage

Custom Model Loading

from openvideo.models.text2video import TextToVideoModel

model = TextToVideoModel(model_name="zeroscope")
model.load()
frames = model.generate(prompt="your prompt")

Batch Generation

prompts = [
    "a dog running in park",
    "a bird flying in sky",
    "rain falling on window",
]

for i, prompt in enumerate(prompts):
    frames = generator.generate(prompt=prompt)
    generator.save_video(frames, f"video_{i}.mp4")

Troubleshooting

Out of Memory

• Reduce resolution (width/height)
• Use float16 dtype
• Reduce num_frames
• Enable attention slicing

Slow Generation

• Use GPU instead of CPU
• Reduce num_inference_steps
• Use smaller models

License

MIT License

Credits

• Uses Hugging Face Diffusers
• Model weights from various open-source projects