zeroscope_v2_576w
| Property | Value |
|---|---|
| License | CC-BY-NC-4.0 |
| Author | cerspense |
| Downloads | 82,736 |
| Training Data | 9,923 clips, 29,769 tagged frames |
What is zeroscope_v2_576w?
zeroscope_v2_576w is a specialized text-to-video generation model designed to create high-quality 16:9 video compositions at 576x320 resolution. This model is particularly notable for producing watermark-free outputs and is optimized for smooth video generation. It's built upon the ModelScope framework and has been specifically trained using nearly 10,000 video clips.
Implementation Details
The model requires 7.9GB of VRAM for generating 30 frames at 576x320 resolution. It's implemented using the Diffusers library and can be easily integrated into existing pipelines. The model is particularly designed to work in conjunction with zeroscope_v2_XL for high-resolution upscaling.
- Built on ModelScope architecture
- Optimized for 24fps video generation
- Supports DPMSolverMultistepScheduler
- Compatible with Diffusers pipeline
Core Capabilities
- Generates smooth, high-quality video outputs
- Produces watermark-free content
- Optimized for 16:9 aspect ratio
- Supports upscaling integration
- Efficient memory usage with model CPU offloading
Frequently Asked Questions
Q: What makes this model unique?
This model's unique strength lies in its optimization for 16:9 compositions and its ability to produce watermark-free outputs. It's specifically designed as a preliminary step for upscaling with zeroscope_v2_XL, allowing for efficient workflow in video generation.
Q: What are the recommended use cases?
The model is best suited for generating initial video content at 576x320 resolution that will be upscaled to higher resolutions. It's particularly effective when used in conjunction with the zeroscope_v2_XL model for upscaling to 1024x576 resolution with a denoise strength between 0.66 and 0.85.
