Skip to content

Latest commit

 

History

History
49 lines (38 loc) · 2.46 KB

README.md

File metadata and controls

49 lines (38 loc) · 2.46 KB
Raw

Upscayl - Custom Models

This repository will contain extra models that users can download and use in Upscayl. Upscayl v2.5 comes with a Custom Models option where users can select a models folder and use their own imported models.

Instructions

Using Models

  1. Download this repository
  2. Extract the downloaded ZIP file.
  3. Open Upscayl and click the Settings tab.
  4. Click on the "Select Folder" button under the "ADD CUSTOM MODELS" section.
  5. Select the models folder in the earlier extracted folder.
  6. Go back to Upscayl screen and select your custom models.

Convert your own Models!

The Upscayl wiki has a guide on how to convert pre-existing ESRGAN models!

Models

General

  • 4x_NMKD-Siax_200k - NMKD Siax - Universal upscaler for clean and slightly compressed images (JPEG quality 75 or better), based on CX loss + PatchGAN.
  • 4x_NMKD-Superscale-SP_178000_G - NMKD Superscale - Perfect upscaling of clean (artifact-free) real-world images.
  • RealESRGANv3 - Lightweight and faster versions of the default model, with (very slightly) worse quality.
    • RealESRGAN_General_WDN_x4_v3 - wide and deep network model
    • RealESRGAN_General_x4_v3
  • unknown-2.0.1 - "Unknown" - We still don't know what this model is, but we accidently included it in place of Ultrasharp in v2.0.1 and @royal-rigolo liked it, so here it is!
  • uniscale_restore by Kim2091.
  • 4xLSDIR by Phhofm.
  • 4xLSDIRplusC by Phhofm.
  • 4xLSDIRCompactC3 as a SRVGGNET (Compact) model, meaning inference times should be faster, by Phhofm.
  • 4xNomos8kSC by Phhofm.
  • 4xHFA2k by Phhofm.

Digital Art

animevideo models for scales x2, x3 and x4.

  • realesr-animevideov3-x2
  • realesr-animevideov3-x3
  • realesr-animevideov3-x4

Note

Upscayl will use the model file name to detect the scale that it supports. If the model file name has 'x2' or '2x' in its name, Upscayl-NCNN will set the target resolution for upscaling accordingly. This doesn't affect the post-processing scaling option, this only affects the actual upscaling via ESRGAN.

Credits:

@xinntao LICENSE
@Kim2091
NMKD @Phhofm