zeroscope_v2_XL
| Property | Value |
|---|---|
| License | CC-BY-NC-4.0 |
| Pipeline Type | Video-to-Video |
| VRAM Usage | 15.3GB (30 frames at 1024x576) |
| Training Data | 9,923 clips, 29,769 tagged frames |
What is zeroscope_v2_XL?
zeroscope_v2_XL is an advanced video generation model designed specifically for upscaling content created with zeroscope_v2_576w. Built on ModelScope architecture, it's trained from original weights using offset noise and specializes in producing high-quality video output at 1024x576 resolution without watermarks.
Implementation Details
The model leverages the Diffusers pipeline and requires specific implementation steps for optimal performance. It's designed to work seamlessly with the 1111 text2video extension and supports various rendering configurations.
- Trained on 24-frame sequences at 1024x576 resolution
- Implements DPMSolverMultistepScheduler for efficient processing
- Supports CPU offloading and VAE slicing for memory optimization
- Compatible with torch float16 for accelerated processing
Core Capabilities
- High-resolution video generation (1024x576)
- Watermark-free output
- Efficient upscaling from lower resolution inputs
- Supports batch processing of multiple frames
- Optimized for 24+ frame sequences
Frequently Asked Questions
Q: What makes this model unique?
The model's specialization in upscaling content from zeroscope_v2_576w makes it unique, allowing for superior compositions at higher resolutions while maintaining quality and efficiency.
Q: What are the recommended use cases?
The model is best suited for upscaling pre-generated videos from zeroscope_v2_576w, with recommended denoise strength between 0.66 and 0.85, and maintaining the same prompt used in the original generation.
