4x photo upscaler for faces with otf jpg compression, blur and resize, trained on FFHQ dataset. This has been trained on and for faces, but i guess can also be used for other photos, might be able to retain skin detail. This is not face restoration, but simply a 4x upscaler trained on faces, therefore input images need to be of good quality if good output quality is desired.

4xFFHQDAT

4x anime image upscaler with a bit of otf jpg compression and blur. Trained on musl's hfa2k dataset release for Anime SISR, which has been extracted from modern anime films, where the selection criteria was high SNR, no DOF and high frequency information.

Examples: https://imgsli.com/MTc2NDgx (PS Example input images are small, each around 500x280 px) \
Example input files: https://drive.google.com/drive/folders/1RI6gGqRy-KxDujbaIrpEMdvFkIWHvfPt \
Example output files: https://drive.google.com/drive/folders/1GqHwPlFp6bIQl4R1AxmrmJ6vUUD8FUxH \
All my model files can be found on my github repo https://github.com/Phhofm/models

HFA2k

[Link to Github Release](https://github.com/Phhofm/models/releases/tag/4xHFA2k_ludvae_realplksr_dysample)

4xHFA2k_ludvae_realplksr_dysample  
Scale: 4  
Architecture: [RealPLKSR with Dysample](https://github.com/muslll/neosr/?tab=readme-ov-file#supported-archs)  
Architecture Option: [realplksr](https://github.com/muslll/neosr/blob/master/neosr/archs/realplksr_arch.py)  

Author: Philip Hofmann  
License: CC-BY-0.4 
Purpose: Restoration  
Subject: Anime  
Input Type: Images  
Release Date: 13.07.2024  

Dataset: HFA2k_LUDVAE  
Dataset Size: 10'272  
OTF (on the fly augmentations): No  
Pretrained Model: [4xNomos2_realplksr_dysample](https://github.com/Phhofm/models/releases/tag/4xNomos2_realplksr_dysample)  
Iterations: 165'000  
Batch Size: 12  
GT Size: 256  

Description:  
A Dysample RealPLKSR 4x upscaling model for anime single-image resolution.  
The dataset has been degraded using DM600_LUDVAE, for more realistic noise/compression. Downscaling algorithms used were imagemagick box, triangle, catrom, lanczos and mitchell. Blurs applied were gaussian, box and lens blur (using chaiNNer). Some images were further compressed using -quality 75-92. Down-up was applied to roughly 10% of the dataset (5 to 15% variation in size). Degradations orders were shuffled, to give as many variations as possible.  

Examples are inferenced with [neosr](https://github.com/muslll/neosr) testscript and the released pth file. I include the test images also as a zip file in this release together with the model outputs, so others can test their models against these test images aswell to compare.  

onnx conversions are  static since dysample doesnt allow dynamic conversion, I tested the conversions with [chaiNNer](https://github.com/chaiNNer-org/chaiNNer).  

Showcase:  
[Slowpics](https://slow.pics/c/FKDZAcyI)

4xHFA2k_ludvae_realplksr_dysample

Photos/realistic 3D with JPEG compression, quality 85-95 and 4:2:0 chroma subsampling. - Created for Myst3 images, since all have 4:2:0 chroma subsampling and existing JPEG models did not give good results. Favors smoothing over over-sharpening.

Jaypeg90

Upscale small photos with compression to 4x their size.

Trying to extend my previous model to be able to handle compression (JPG 100-30) by manually altering the training dataset, since 4xLSDIRCompact can't handle compression. Use this instead of 4xLSDIRCompact if your photo has compression (like an image from the web).

LSDIR Compact C

Upscale compressed photos to x4 their size.

Able to handle JPG compression (30-100).

LSDIR Compact C3

A 4x photo upscale DAT model trained with otf (resize, jpg, small blur) on the LSDIR dataset.

4xLSDIRDAT

Interpolation of 4xLSDIRplusC and 4xLSDIRplusR to handle jpg compression and a little bit of noise/blur

4xLSDIRplus

The RealESRGAN_x4plus finetuned with the big LSDIR dataset (84,991 images / 165 GB), with manually added jpg compression.

4xLSDIRplusC

The RealESRGAN_x4plus finetuned with the big LSDIR dataset (84,991 images / 165 GB), with jpg compression and noise and blur

4xLSDIRplusR

Purpose: Art/CGI

Upscaler for clean images or images with compression artifacts (jpeg quality >75) - Produces very sharp lines/edges due to NN-Filtered HR images. Proven to produce very, very good results on drawings (sharp lines) and CGI, but should also work pretty well for real-world images.

NMKD PatchySharp

Universal upscaler for clean and slightly compressed images (JPEG quality 75 or better)

NMKD Siax ("CX")

4xNomos2_otf_esrgan

Scale: 4  
Architecture: ESRGAN  
Architecture Option: esrgan  
[Github Release Link](https://github.com/Phhofm/models/releases/tag/4xNomos2_otf_esrgan)  

Author: Philip Hofmann  
License: CC-BY-0.4  
Purpose: Restoration  
Subject: Photography  
Input Type: Images  
Release Date: 22.06.2024  

Dataset: Nomos-v2  
Dataset Size: 6000  
OTF (on the fly augmentations): Yes  
Pretrained Model: RealESRGAN_x4plus  
Iterations: 246'000  
Batch Size: 8  
GT Size: 256  

Description:  

4x ESRGAN model for photography, trained using the Real-ESRGAN otf degradation pipeline.  

Showcase:
[Slow Pics 8 Examples](https://slow.pics/s/Y5wdyTPh?canvas-mode=fit-width&image-fit=contain)

4xNomos2_otf_esrgan

4xNomos2_realplksr_dysample  
Scale: 4  
Architecture: [RealPLKSR with Dysample](https://github.com/muslll/neosr/?tab=readme-ov-file#supported-archs)  
Architecture Option: [realplksr](https://github.com/muslll/neosr/blob/master/neosr/archs/realplksr_arch.py)  
[Github Release](https://github.com/Phhofm/models/releases/tag/4xNomos2_realplksr_dysample)  

Author: Philip Hofmann  
License: CC-BY-0.4  
Purpose: Pretrained  
Subject: Photography  
Input Type: Images  
Release Date: 30.06.2024  

Dataset: [nomosv2](https://github.com/muslll/neosr/?tab=readme-ov-file#-datasets)  
Dataset Size: 6000  
OTF (on the fly augmentations): No  
Pretrained Model: [4xmssim_realplksr_dysample_pretrain](https://github.com/Phhofm/models/releases/tag/4xmssim_realplksr_dysample_pretrain)  
Iterations: 185'000  
Batch Size: 8  
GT Size: 256, 512  

Description: 
A Dysample RealPLKSR 4x upscaling model that was trained with / handles jpg compression down to 70 on the Nomosv2 dataset, preserves DoF.   
Based on the [4xmssim_realplksr_dysample_pretrain](https://github.com/Phhofm/models/releases/tag/4xmssim_realplksr_dysample_pretrain) I released 3 days ago.  
This model affects / saturate colors, which can be counteracted a bit by using wavelet color fix, as used in these examples.  

Showcase:
[Slowpics](https://slow.pics/s/p3sMnd5l)

4xNomos2_realplksr_dysample

4x photo upscaler with otf jpg compression and blur, trained on musl's Nomos8k_sfw dataset for realisic sr

Nomos8kSC

Purpose: 4x universal DoF preserving upscaler

4x universal DoF preserving upscaler, pair trained with jpg degradation (down to 40) and multiscale (down_up, bicubic, bilinear, box, nearest, lanczos) in neosr with adamw, unet and pixel, perceptual, gan and color losses.
Similiar to the last model I released, with same dataset, this is a full RGT model in comparison.

FP32 ONNX conversion is provided in the google drive folder for you to run it.

6 Examples (To check JPG compression handling see Example Nr.4, to check Depth of Field handlin see Example Nr.1 & Nr.6):
[Slowpics](https://slow.pics/s/iMuE3vE5)

NomosUni rgt multijpg

Purpose: 4x universal DoF preserving upscaler

4x universal DoF preserving upscaler, pair trained with jpg degradation (down to 40) and multiscale (down_up, bicubic, bilinear, box, nearest, lanczos) in neosr with adamw, unet and pixel, perceptual, gan, color and ldl losses.

Examples:
[Imgsli](https://imgsli.com/MjQwNDk5)
[Slowpics](https://slow.pics/s/sIRl3w7J)

NomosUni rgt s multijpg

4x fast universal upscaler pair trained with jpg degradation (down to 40) and multiscale (down_up, bicubic, bilinear, box, nearest, lanczos).

4x-NomosUni_span_multijpg

Purpose: 4x universal DoF preserving upscaler

4x universal DoF preserving upscaler, pair trained with jpg degradation (down to 40) and multiscale (down_up, bicubic, bilinear, box, nearest, lanczos).

NomosUniDAT2 multijpg ldl

NomosUniDAT2 multijpg ldl sharp

Category: Images
Purpose: Stylization

Photo stylization from JPEGs. Trained on images upscaled by ISO Denoise v2 -> DeJPEG Fatality PlusULTRA -> NMKD Yan2. Essentially a combination of that model chain into one. Use if you're looking for a stylized output, not photo quality.

Pretrained model: NMKD Yandere2

Rek's Effeks Photoanime v2

Description: 4x photo upscaler on the SSDIR_Sharp dataset, trained with otf for jpg, resize and small blur. This model sharpens.

4xSSDIRDAT

## 4xTextures_GTAV_rgt-s_dither

**Scale:** 4  
**Architecture:** [RGT](https://github.com/zhengchen1999/RGT)  
**Architecture Option:** RGT-S  

**Author:** Philip Hofmann  
**License:** CC-BY-0.4  
**Purpose:** Restoration  
**Subject:** Game Textures  
**Input Type:** Images  
**Release Date:** 08.05.2024  

**Dataset:** [GTAV_512_Textures](https://discord.com/channels/547949405949657098/905446120333930566/1132367991808466975)  
**Dataset Size:** 7061  
**OTF (on the fly augmentations):** No  
**Pretrained Model:** 4xTextures_GTAV_rgt-s  
**Iterations:** 128'000  
**Batch Size:** 6,4  
**GT Size:** 128,256  

**Description:**
A model to upscale game textures, trained on GTAV Textures, handles jpg compression down to 80 and was trained with dithering. Basically the previous 4xTextures_GTAV_rgt-s model but extended to handle dithering.

**Showcase:**  
[Slow Pics 25 Examples](https://slow.pics/s/EW7Ifiuw)

4xTextures_GTAV_rgt-s_dither

Pretrained: 4xESRGAN

This is my best model yet! It generates lots and lots of detail and leaves a nice texture on images. It works on most images, whether compressed or not. It does work best on JPEG compression though, as that's mostly what it was trained on. It has the ability to restore highly compressed images as well! If you want a more balanced output, check out the UltraMix Collection down below. It's a bunch of interpolated models based around UltraSharp and my other models

UltraSharp

Vast improvement over v1 in low frequency detail; moiré and artifacting reduced significantly and less random noise from JPEG artefacts in the input. Also now only works on 1 channel images, so it runs slightly faster on average and resulting images are much smaller but it might not work on some ESRGAN implementations, I personally recommend using chaiNNer. v1 may still be better in some edge cases.

There's also a supplementary 1x model that denoises very low quality LRs and smooths halftones so the image works better with the 4x model. Only trained it to help build the dataset and it's useless for already decent-ish LRs but may help you in some situations.

Architecture	RGT
Scale	4x
Color Mode	RGB
License	CC-BY-4.0 Private use Commercial use Distribution Modifications Credit required State Changes No Liability & Warranty Disclaimer
Date	2024-02-20
Dataset	nomosuni
Dataset size	2989
Training iterations	100000
Training epochs	79
Training batch size	12
Training HR size	128
Training OTF	No

NomosUni rgt multijpg

Similar Models