| Overview |
It is the upgraded model that is faster than V2. Generate more details. Deliver more clarity and sharpness. Improve the details of skin, hair, and overall perceptual quality. |
It is another upgrade AI model trained to restore imperfect images and eliminate flaws for better fidelity, enhanced clarity, and a smoother appearance. |
It is a restoration model made for anime videos and images. It is trained using millions of anime data. |
It aims at developing Practical Algorithms for general video/image restoration, trained with pure synthetic data. |
| Best for |
Enhancing media with rich textures and details; generating details and sharpening images; fixing blurry pictures; producing realistic and detail-rich outputs. |
Removing noise, restoring flaws, and upscaling images. Maintaining the original artistic style, coherence, or natural appearance. Producing visually pleasing outputs. |
Upscaling anime while maintaining realistic content and a natural appearance. Usable on average computers. |
Enhancing image quality by estimating image kernels and injecting noise. Usable on average computers. |
| Recommended Media Types |
Low-quality or low-resolution videos, digital images, AI art, compressed JPEGs; videos/images with skin, hair, pores, textures, or intricate patterns. |
Images/videos that do not need more details, such as some human faces, smooth surfaces, animations, or line drawings. |
|
- Reality videos and images that do not need too many details.
|
| Scaling/Output |
1x, 2x, 3x, 4x; 480p, 720p, 1080p, 2K, 4K |
1x, 2x, 3x, 4x; 480p, 720p, 1080p, 2K, 4K |
1x, 2x, 3x, 4x; 480p, 720p, 1080p, 2K, 4K |
1x, 2x, 3x, 4x; 480p, 720p, 1080p, 2K, 4K |
| Overall Clarity |
|
|
|
|
| Denoise |
|
Best for most videos.
|
Only suitable for Anime.
|
It will smooth out details.
|
| Deblur |
|
|
|
|
| Sharpness |
|
|
|
|
| More Details |
|
|
|
|
| Modes |
Fast Mode; High Quality Mode |
Fast Mode; High Quality Mode |
Fast Mode; High Quality Mode |
Fast Mode; High Quality Mode |
| Supported OS |
Windows: Microsoft Windows 10 (x64) 1809 or higher;
Mac: macOS 10.15 or later. |
Windows: Microsoft Windows 10 (x64) 1809 or higher; Mac: macOS 10.15 or later. |
Windows: Microsoft Windows 7 or higher;
Mac: macOS 10.5 or later. |
Windows: Microsoft Windows 7 or higher; Mac: macOS 10.5 or later. |
| CPU/GPU Requirement |
CPU, GPU
• NVIDIA: GTX Titan X, GTX 980 Ti, GTX 980, GTX 970, GTX 960 or higher;
• Intel: Haswell (4th-gen core) HD Integrated Graphics or higher
• AMD: Radeon R9 390(X), Radeon R9 395X2 or higher.
Meet the minimum requirements of DirectML or TensorRT hardware acceleration.
|
CPU, GPU
• NVIDIA: GTX Titan X, GTX 980 Ti, GTX 980, GTX 970, GTX 960 or higher;
• Intel: Haswell (4th-gen core) HD Integrated Graphics or higher
• AMD: Radeon R9 390(X), Radeon R9 395X2 or higher.
Meet the minimum requirements of DirectML or TensorRT hardware acceleration.
|
CPU, GPU
• NVIDIA: GeForce GT 630 or higher;
• Intel: HD Graphics 2000 or higher;
• AMD: Radeon HD 7700 series (VCE 1.0) or higher
Vulkan ≥ 1.2.X
|
CPU, GPU
• NVIDIA: GeForce GT 630 or higher;
• Intel: HD Graphics 2000 or higher;
• AMD: Radeon HD 7700 series (VCE 1.0) or higher
Vulkan ≥ 1.2.X
|
| Architecture |
GAN (Generative Adversarial Network) |
Diffusion |
GAN (Generative Adversarial Network) |
Super-Resolution Model with kernel estimation and noise injection |