Model Database

If a model has no license, that means all rights reserved. You will need a private license or explicit permission from the creator to use it commercially or with modifications. However you may use all model here on non-commercial projects, that may not in any way cost money. If you gain permission to use a model in another way including commercially, you should still credit the author and link to the original model source.

ESRGAN ("old Architecture") Models

Models that use the "old" ESRGAN architecture. They can be used either with the official ESRGAN repository (old arch tag) and BasicSR (old arch or preferably victorc's fork). You can also use tools that wrap around one of them, like IEU from Honh.

Image scaling and Video upscaling

In computer graphics and digital imaging, image scaling refers to the resizing of a digital image. In video technology, the magnification of digital material is known as upscaling or resolution enhancement.

Drawings

Drawing is a form of visual art in which a person uses various drawing instruments to mark paper or another two-dimensional medium. Instruments include graphite pencils, pen and ink, various kinds of paints, inked brushes, colored pencils, crayons, charcoal, chalk, pastels, various kinds of erasers, markers, styluses, and various metals (such as silverpoint). Digital drawing is the act of using a computer to draw. Common methods of digital drawing include a stylus or finger on a touchscreen device, stylus- or finger-to-touchpad, or in some cases, a mouse. There are many digital art programs and devices.

Manga/Anime

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
Manga109Attempt	Kingdomakrillic	4	CC BY 4.0	Anime / Manga	?	4	?	?	0.1K	Manga109	RRDB_PSNR_x4
Falcon Fanart	LyonHrt	4	CC BY-NC-SA 4.0	Anime / Manga	125K	8	128	?	3.393K	Falcon Fanart	RRDB_PSNR_x4
WaifuGAN v3	DinJerr	4	CC BY-NC 4.0	Anime / Manga	30K	2	128	?	0.173K	CG-Painted Anime	Manga109v2
De-Toon	LyonHrt	4	CC BY-NC-SA 4.0	Toon Shading / Sprite	225K	8	128	0.525K	7.117K	Custom Cartoon-style photos	RRDB_PSNR_x4
DigiPaint	TheAtheistGod	4	CC0	Digital Art Upscaler	35K	6	128	6	48.493K	Misc digital art, material studies	4xfalcoon300(manga)
DeviantPixelHD	Raulsangonzalo	4	CC BY-NC 4.0	Digital Art Upscaler	250K	16	128	1.327K	2.797K	Digital Art from Deviant Art	RRDB_PSNR_x4
FireAlpha	BlueAmulet	4	CC BY-NC 4.0	Artwork with an alpha channel	1.285M		128		104k tiles, 1.2k imgs	Fire Emblem artwork with transparency	None

Manga109Attempt is slightly blurry, but performs well as a general upscaler.

Falcoon Fanart tries to improve upon it with the goal of removing checkerboard patterns / and dithering. It has oil color based shading with sharp lines.

WaifuGAN v3 is DinJerr’s third attempt at training from a mostly anime dataset sourced from image boards and is intended for upscaling CG-painted anime with variable outlines. Only PNGs were used, mainly with brush strokes and gradients. Texturised images avoided as much as possible. If too generative, tone down by interpolating with a softer model.

De-Toon, is a model that does the opposite of tooning an image. It takes toon style shading and detail, and attempts to make it realistic. Its very sensitive, and can be used on small sprites, to large images. Also included is a alt version, which is less sharp.

DigiPaint is a digital art upscaler designed to take brush strokes into account, based off of the Falcoon Fanart model.

DeviantPixelHD was designed to upscale original LucasArts games' backgrounds. High Definition digital art from Deviant Art was used, reducing the LR training to 32 colors so the training images looked pixelated. Can be used with other environments such as manga or digital art. Examples can be found on Raúl Sangonzalo's YouTube account (https://www.youtube.com/channel/UCwBfuiHdSPQ-zslOmiW8OHg).

FireAlpha is an attempt to create a 4 channel model that handles transparency without splitting and upscaling separately.

• Notice: IEU can use the model but will attempt to split and handle transparency itself, BlueAmulet's fork of ESRGAN is required to use this model. https://github.com/BlueAmulet/ESRGAN

Cartoon / Comic

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
Comic Book	LyonHrt	4	CC BY-NC-SA 4.0	Comic / Drawings	115K	8	128	592	1.548K	Custom (Spider-Man)	none (no interpolation)
DigitalFrames 1.0	Klexos	4	CC BY-NC-SA 4.0	Clean Digital Animation	1.06M	15	128	96.275	0.25K - 2.5K	Digital Cartoon Images	RRDB_PSNR_x4
DigitalFrames 2.0	Klexos	4	CC BY-NC-SA 4.0	Digital Animation	905K	27	128	7	1.636M	Digital Cartoon Images	RRDB_PSNR_x4
DigitalFrames 2.1_Aggressive	Klexos	4	CC BY-NC-SA 4.0	Flat Digital Animation	230K	27	128	2	1.8M	Digital Cartoon Images	DigitalFrames 2.0
DigitalFrames 2.1_Final	Klexos	4	CC BY-NC-SA 4.0	Digital & Traditional Animation	230K	27	128	2	1.8M	Cartoon Images	DigitalFrames 2.0
FatalimiX	Twittman	4	CC BY-SA 4.0	Comic and Cartoon style images	260k	10	128	31	69k/79k	Digital Comics	4x_Fatality_MKII_90000_G_02.pth
SpongeBob	Joey	4	CC BY-NC-SA 4.0	Cartoons & SpongeBob Games	235k	20	128	585	8k	Scene-detected frames from random Season 11 episodes	4xESRGAN
SpongeBob v6	Joey	4	CC BY-NC-SA 4.0	Cartoons	190k	20	128		4,803	1 frame of every scene of season 11, downscaled 50% with nearest neighbor	4xESRGAN
Spongebob v6 De-Quantize	Joey	4	CC BY-NC-SA 4.0	Quantized Cartoons	90k	20	128		4,803	1 frame of every scene of season 11, downscaled 50% with nearest neighbor	4x Spongebob v6
Spongebob v6 Deblur	Joey	4	CC BY-NC-SA 4.0	Cartoons	65k	20	128		4,803	1 frame of every scene of season 11, downscaled 50% with nearest neighbor	4x Spongebob v6
HugePeeps	DinJerr	8	CC BY-NC 4.0	Painted humans	360k	4	256		1,000+	A bunch of pretty girls (and guys) from ArtStation	TGHQFace8x
Fatal_Anime	Twittman	4	CC BY-SA 4.0	Anime and Cartoons	500k	6	128	18	183k	Anime and Manga images	4x_Fatality_Comix_MKII_430000_G.pth
HugePaint	DinJerr	8	CC BY-NC 4.0	Digital Illustrations	500k	4	256		1,000+	Variety of images from ArtStation	HugePeeps

The Comic Book model was trained using stills from the film spiderman into the spiderverse, has a comic book crosshatch shading effect to the images. Sample

DigitalFrames 1.0: The purpose of this model is upscaling digital *only* animation frames while cleaning and restoring textures and details, but it can also works well on a variety of other things such as medium sized sprites and digital art. This model is highly sensitive to grain and other type of noise so if an improper input image is feed, the results can have severe blobs of noise if the source image has any type of grain, even if barely noticeable.

DigitalFrames 2.0: Was trained completely separately from 1.0 with a completely new and bigger dataset. It has a reduced sensitivity to noise which reduces the blobs of noise phenomena. It can also work reasonably well on Cel/Film Toons now. Overall it's a big leap from 1.0.

DigitalFrames 2.1_Aggressive: Provides the sharpest results yet. About 30% better than the base 2.0 version. But as the title says it's the most aggressive version of DF. Does it's best on animation with little to no textures. The model as well as any version of DF is not suitable at all for overly deteriorated/blurred and low quality TVRips and VHSRips copies of animation pieces.

DigitalFrames 2.1_Final: The most universally "compatible" from the 2.0 branch that works well for all types of animation. If your piece has few textures or none at all, 2.1_Aggressive can perform better, but for all other scenarios _Final takes the cake with a much higher texture retention and "natural" look.

FatalimiX is specialized in making those low res comics high res again!

The SpongeBob model was trained for the purpose of upscaling Battle for Bikini Bottom's textures, but it works well for upscaling scenes from the show as well. It was initially trained on the clean downscaled images but later JPEG, quantization, dithering, and Gaussian blur were added to the LR OTF options to help it upscale the textures better.

Spongebob v6 is ideally a lot sharper and cleaner, but Joey's still not sure if it works better than the old one. It still seems to be better in many cases.

Spongebob v6 De-Quantize is for de-quantizing as well as upscaling. The results are pretty blurry but it works decently enough.

Spongebob v6 Deblur was created as Joey wasn't entirely happy with the main Spongebob v6 model. It was trained with blurring OTF options and two different downscale types.

HugePeeps v1 was created by DinJerr to test out a theory on how to properly train 8x models. This particular model best works on painted pics of ladies.

HugePaint v1 is trained on a more generic dataset with a balance of illustrated humans, vegetation, minerals and some animals.

Some of the first cartoon painted models for ESRGAN can still be found here, but have been mostly superseded by newer models.

Pixel Art

Pixel art is a form of digital art, created through the use of software, where images are edited on the pixel level. The aesthetic for this kind of graphics comes from 8-bit and 16-bit computers and video game consoles, in addition to other limited systems such as graphing calculators. In most pixel art, the color palette used is extremely limited in size, with some pixel art using only two colors.

Creating or modifying pixel art characters or objects for video games is sometimes called spriting, a term that arose from the hobbyist community. The term likely came from the term sprite, a term used in computer graphics to describe a two-dimensional bitmap that is used in tandem with other bitmaps to construct a larger scene.

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
Xbrz	LyonHrt	4	CC BY-NC-SA 4.0	Xbrz style pixel art upscaler	90K	8	128	368	1.897K	custom xbrz up-scaled	RRDB_PSNR_x4
Xbrz+DD	LyonHrt	4	CC BY-NC-SA 4.0	Xbrz style pixel art upscaler with de-dithering	90K	8	128	470	1.523K	custom de-dithered xbrz	xbrz
ScaleNX	LyonHrt	4	CC BY-NC-SA 4.0	Scalenx style pixel art upscaler	80K	8	128	599	1.070K	custom scalenx up-scaled from retroarch shader	RRDB_PSNR_x4
Fatality	Twittman	4	CC BY-SA 4.0	(dithered) spirites	265K	10	128	160	19.7K	?	Face
Rebout	LyonHrt	4	CC BY-NC-SA 4.0	Character Sprites	325K	8	128	106	23.808K	Custom prepared sprites from kof 94 rebout	Detoon
Lady0101	DinJerr	4	CC BY-NC 4.0	Painting-style	340K	4	128	52	7K	CG-Painted Pinups & Landscapes	WaifuGAN v3
Faithful 2x	Joey	2	CC0	General small pixel textures	130K	80	32	3713	2.858K	Faithful 32x32 (HR) / default Minecraft 1.13 and 1.14 (LR)	Private model trained from 2xESRGAN
FArtDIV3 Suite	DinJerr	4	CC BY-NC 4.0	Pixels to digital painting	700K~	4	128	100~	10K~	DIV2K, FatalityFaces, ArtStation paintings	ArtStation1337
BigFArt Suite	DinJerr	4	CC BY-NC 4.0	Larger-scaled pixels to digital painting	800+	6	192	200~	12K~	DIV2K, FatalityFaces, ArtStation paintings	Face Focus
Fatal Pixels	Twittman	4	CC BY-SA 4.0	Sprites	340K	22	96	151	49K	Anime, Manga	Fatality_MKII_90k
SmoothRealism	Joey	4	CC BY-NC 4.0	Pixel art, rocky/grainy textures	140K	2	64	?	679	R3DCraft Smooth Realism Texture Pack 64x	RRDB_PSNR_x4

Fatality is meant to be used for upscaling medium resolution Sprites, dithered or un-dithered, it can also upscale manga/anime and gameboy camera images.

Rebout is trained to give detail to character models, with faces and hands improved. Based on the snk game kof94 rebout, although best for snk style games, does work on a variety of sprites. Also included is a interpolated version that may provide a cleaner upscale for certain sprites.

Lady0101 was trained on digital paintings (mostly of pinups) and landscapes. It is meant to upscale pixel art into painting-style images with some canvas-like texturing for certain elements. Included is a version that has strong blending but weak undithering.

Faithful 2x was trained on the faithful 32x32 Minecraft texture/resource pack, using both the 1.13 and 1.14 versions (which have different textures). For those not familiar, this texture pack is meant to upscale the 16x16 textures to 32x32 faithfully, as in make them as close as possible to the original, but higher resolution. The model does a decent job at replicating this.

• NOTE: Due to an error in the dataset, sometimes the model turns jagged white lines green. Only encountered this in one texture tried, so it appears unlikely to happen.

FArtDIV3 Suite is a group of models based on a large image dataset trained using OTF tile generation. Most of the models are based off the Base model, and suite different purposes. Fine is used to generate details, but may be noisy/grainy at times. Blend will attempt to blend colour steps into proper gradients. UnholyFArt is a interpolated model that reduces details in favor of sharp blends.

BigFArt Suite is similar to FArtDIV3, but for larger pixel sprites/art in the context of tile size.

Photographs and Photorealism

A photograph (also known as a photo) is an image created by light falling on a photosensitive surface, usually photographic film or an electronic image sensor, such as a CCD or a CMOS chip. Most photographs are created using a camera, which uses a lens to focus the scene's visible wavelengths of light into a reproduction of what the human eye would see. The process and practice of creating such images is called photography. The word photograph was coined in 1839 by Sir John Herschel and is based on the Greek φῶς (phos), meaning "light," and γραφή (graphê), meaning "drawing, writing," together meaning "drawing with light." Photorealism is a genre of art that encompasses painting, drawing and other graphic media, in which an artist studies a photograph and then attempts to reproduce the image as realistically as possible in another medium. Although the term can be used broadly to describe artworks in many different media, it is also used to refer specifically to a group of paintings and painters of the American art movement that began in the late 1960s and early 1970s.

Misc / Kitchen Sink

All kinds of photographs or photorealistic images. Those models aren't specialized.

Box was meant to be an improvement on the RRDB_ESRGAN_x4 model (comparison). It’s also trained on photos, but with a much larger dataset which was downscaled with linear interpolation (box filter) instead of bicubic.

Deoldify is intended for denoising and deblurring photos. All outputs are greyscale. This was initially a proof of concept. (Need clarification on specific subject matter.)

The Ground model was trained on various pictures of stones, dirt and grass using Google’s image search.

The Misc model is trained on various pictures shoot by myself, including bricks, stone, dirt, grass, plants, wood, bark, metal and a few others.

Characters and Faces

For images of humans, creatures, faces, ...

Trixie was made to bring balance to the force… Also to upscale character textures for star wars games, including the heroes, rebels, sith and imperial. Plus a few main aliens…Why called trixie? Because jar jars big adventure would be too long of a name…This also provides good upscale for face textures for general purpose as well as basic star wars textures.

The Face Focus modes was designed for slightly out of focus / blurred images of faces. It is aimed at faces / hair, but it can help to improve other out of focused images too as always just try it.

Specialized

For Purposes that didn't fit anywhere else (for now).

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
Map	LyonHrt	4	CC BY-NC-SA 4.0	Map / Old Paper with text	120K	8	192	361	2.311K	Custom(Scans)	none
Forest	LyonHrt	4	CC BY-NC-SA 4.0	Wood / Leaves	160K	8	192	590	2.2K	Custom(?)	none
Skyrim Armory	Alsa	4	CC BY-NC-SA 4.0	Armor, Clothes and Weapons	80K	26	128	2.6K	0.8K	Skyrim Mod textures	Manga109Attempt
Skyrim Wood	Laeris	4		Wood	75K	?	?	?	?	?	?
Skyrim Misc	Deorder	4	CC0	Skyrim Diffuse Textures	105K	?	128	?	Skyrim Diffuse Textures	?
Fallout 4 Weapons	Bob	4	CC BY-NC-SA 4.0	Fallout Weapon Diffuse Textures	120K	13	128	2.973K	532(OTF)	Fallout 4 HDDLC Weapon Diffuse Textures	Manga109Attempt
Fallout Weapons V2	Bob	4	CC BY-NC-SA 4.0	Video game textures, mostly metal rusty, clean or painted.	180K	13	128	1.19K	1.999K(OTF)	Fallout 4 HD DLC(Weapon,Armor/Clothes,Vehicle and Interior/Architecture textures to be exact)	Fallout 4 Weapons
Nickelfront	BlackScout	4	GNU GPLv3	Money, Coins	14k	8	128	105	1.176k	Images of Coins	4xESRGAN
BS_Wolly	BlackScout	2	GNU GPLv3	Pixar Movies / Wall-E	38k	1	256	3	9.0k	BD 1080p Wall-E Frames	2xESRGAN

The map model was trained on maps, old documents, papers and various styles of typefaces/fonts. Based on a dataset contributed by alsa. Sample

The Forest model is focused on trees, leaves, bark and stone can be used for double upscaling for even more detail. Sample

The Armory model was trained with modded textures form Skyrim, including Clothing, Armor and Weapons. (Leather, Canvas and Metal should all work - maybe too sharp so interpolate)

The wood model was trained for Skyrim by Laeris.

The Skyrim Diffuse models is supposed to be used with Skyrim’s diffuse textures. It is a bit too sharp so I recommend to interpolating with the RDDB_ESRGAN_x4 model or the mangaAttempt109 model, look in Deorder’s Skyrim Model Google Drive for an already interpolated version.

Fallout 4 weapons was trained using Fallout 4’s official HDarmor/weapon textures but could be used on other weapon and armor textures.

Texture Maps

Normal Maps

In 3D computer graphics, normal mapping, or Dot3 bump mapping, is a technique used for faking the lighting of bumps and dents – an implementation of bump mapping. It is used to add details without using more polygons. A common use of this technique is to greatly enhance the appearance and details of a low polygon model by generating a normal map from a high polygon model or height map.

Normal maps are commonly stored as regular RGB images where the RGB components correspond to the X, Y, and Z coordinates, respectively, of the surface normal.

The models here have been specifically trained on Normal Maps, but beware, this approach is considered deprecated by many of us. Instead of using any of those models, you can just split the R, G and B channels of the normal map you want to upscale and use any other model on them.

The first one is based on the second one it was trained, with a higher learning rate and insane n_workers and batch_size values. It is meant to replace the old Normal Map model from Deorder, but without adding BC1 compression to your normal maps.

The second one was trained on Skyrim’s Normal Maps, including compression artifacts, so it will have to be redone.

Grayscale

In digital photography, computer-generated imagery, and colorimetry, a grayscale or greyscale image is one in which the value of each pixel is a single sample representing only an amount of light, that is, it carries only intensity information. Grayscale images, a kind of black-and-white or gray monochrome, are composed exclusively of shades of gray. The contrast ranges from black at the weakest intensity to white at the strongest.[1]

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
Skyrim Alpha	Deorder	4	CC0	Alpha Channel	105K	?	128	?	?	Alpha Channels from Skyrim	?
Deoldify	TheAtheistGod	4	CC0	B/W Photo Restoration	790K	16	128	?	2.931M	ADE20K, DIV2K, Library of Congress	Falcon Fanart (?)

Trained to upscale grayscale images, like specular or alpha etc.

Artifact Removal

The models in this section were made to remove compression artifacts in images and textures.

Blur

Fatality DeBlur is intended for general deblurring of images, and has some robustness against compression and noise.

LADDIER1 upscales images with noise, grain and different types of blur.

Noise

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
ISO denoise v1	Alpha	1	CC0	Remove high ISO noise	372K	1	128	2	170K	Custom (DSLR photos)	1xESRGAN.pth
ISO denoise v2	Alpha	1	CC0	Remove high ISO noise	308K	1	128	1	239K	Custom (DSLR photos)	1x_ISO_denoise_v1.pth

High ISO is required to capture images at very high shutter speeds and/or in poor lighting. However, increasing ISO also results in noise that is challenging to remove without also losing fine details. To remove this noise without damaging details, one can apply the technique of median stacking: combine 10-100 photos of the exact same object to separate the signal (clear photo) from noise. ISO denoise v1 is trained on a handcraft set of pairs between original noisy photos and median stacked counterparts. It tends to be conservative, avoiding modifying anything that does not closely resemble camera ISO noise. ISO denoise v2 uses a larger dataset, handling some cases better and some cases worse than v1. If results are unsatisfactory with one model, try the other model.

JPG Compression

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
JPG (0-20%) A	Alsa	1	CC BY-NC-SA 4.0	JPG compressed Images	178K	2	128	52	6.23K	Custom (Photos / Manga)	JPG (20-40%)
JPG (0-20%) B	BlueAmulet	1	CC BY-NC 4.0	JPG compressed Images	610K	1	128	11	52.789K	Custom (CC0 Textures)	JPG (20-40%)
JPG (20-40%) A	Alsa	1	CC BY-NC-SA 4.0	JPG compressed Images	141K	2	128	42	6.23K	Custom (Photos / Manga)	JPG (40-60%)
JPG (20-40%) B	BlueAmulet	1	CC BY-NC 4.0	JPG compressed Images	535K	1	128	10	52.789K	Custom (CC0 Textures)	JPG (40-60%)
JPG (40-60%) A	Alsa	1	CC BY-NC-SA 4.0	JPG compressed Images	100K	2	128	31	~6.5K	Custom (Photos / Manga)	JPG (60-80%)
JPG (40-60%) B	BlueAmulet	1	CC BY-NC 4.0	JPG compressed Images	550K	1	128	10	52.789K	Custom (CC0 Textures)	JPG (60-80%)
JPG (60-80%) A	Alsa	1	CC BY-NC-SA 4.0	JPG compressed Images	91K	2	128	27	~6.5K	Custom (Photos / Manga)	JPG (80-100%)
JPG (60-80%) B	BlueAmulet	1	CC BY-NC 4.0	JPG compressed Images	545K	1	128	10	52.789K	Custom (CC0 Textures)	JPG (80-100%)
JPG (80-100%) A	Alsa	1	CC BY-NC-SA 4.0	JPG compressed Images	162K	2	128	51	~6.5K	Custom (Photos / Manga)	BC1 take 1
JPG (80-100%) B	BlueAmulet	1	CC BY-NC 4.0	JPG compressed Images	550K	1	128	10	52.789K	Custom (CC0 Textures)	BC1 take 1
JPG PlusULTRA	Twittman	1	CC BY-SA 4.0	JPG compressed Images	130K	1	?	150	0.937K	Custom (Manga)	Failed Attempts
DeJpeg Fatality PlusULTRA	Twittman	1	CC BY-SA 4.0	JPG compressed Images	200K	1	128	5	26K	Real life, Manga and digital text	1x_DeJpeg_Fatality_01_200000_G.pth

JPG gets compressed with a Quality Percentage between 0 and 100. So depending on how bad your JPEGs are compressed, choose the model of your choice. You can use ImageMagick to guess the Quality percentage, but keep in mind that it might be wrong, since the image might have been re-saved.

JPG PlusULTRA, while not considered successful in its initial purpose, may find some use on certain unique images.

DeJpeg Fatality PlusULTRA is its burly strongman big brother. Anything and everything that is JPG will tremble before you!

DDS Files with BC1/DXT1, BC3/DXT5 Compression

BC1 (DXT1) compression is commonly used in dds textures, which are utilized in most PC games today, it allows to shrink the texture to 1/6 of the original size, reducing VRAM usage. There is also BC3 (DXT5) which uses BC1 compression for the color channels and leaves the alpha channel uncompressed, this one reduces the file size to 1/3 of the original. But this compression comes at a cost. The BC1 models are designed to reverse the damage done by the compression. This is a must if you want to upscale a dds file that uses either BC1 or BC3 compression. The free variant has more freedom when dealing with the images and should lead to better results. The restricted version only deals with perfect images compressed once (ideal case) so will perform worse in any other scenario, but it tries to preserve the original colors more. You can interpolate between restricted and unrestricted if you want.

Cinepak, msvideo1 and Roq

The Cinepak model removes movie compressions artifacts from older video compression methods like Cinepak, msvideo1 and Roq. Cinepak_alt may work better for video frames with realistic content.

Color Reduction

Dithering

Dithering is an older compression method, where the amount of colors gets reduced, if your image has few colors and have a noise like pattern in them try a De-Dither model. Ordered dithering is a less common form of dithering that results in distinctive checkerboard/crosshatch patterns, which are misinterpreted as texture by models not trained on it. It’s often used on GIFs because the pattern is stable between frames.

Banding

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
BS_DeBandizer	BlackScout	1	GNU GPLv3	Images with Banding	34k	1	128	37	changing ~900	Custom (Wallpapers)	old attempts based on 1xESRGAN

Banding is like dithering, but instead of adding patterns of pixels to simulate missing colors, it just clips them off.

Color Removal / Grayscale

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
BS_Colorizer	BlackScout	1	GNU GPLv3	Coloring B/W images	22k	1	128	0	4.5045k	Custom (800 Wallpapers)	1xESRGAN

Just removes all colors and makes the image grayscale, the models here should restore some color to them, should also work for old black and white images.

Over-Sharpening

The De-Sharpen model was made for rare particular cases when the image was destroyed by applying noise, i.e. game textures or any badly exported photos. If your image does not have any over-sharpening, it won’t hurt them, leaving as is. In theory, this model knows when to activate and when to skip, also can successfully remove artifacts if only some parts of the image are over-sharpened, for example in image consisting of several combined images, 1 of them with sharpen noise. It is made to remove sharpen noise, particularly made with Photoshop “sharpen” or “sharpen more” filters OR ImageMagick’s -sharpen directive with several varying parameters of Radius and Sigma, from subtle 0.3x0.5 to something extreme like 8x2, somewhere about that.

Aliasing

AntiAliasing is for smoothing jagged edges in images and textures.

Image Generation

Texture Maps

The model was trained on pairs of diffuse textures and normal maps.

Pretrained models for different scales

Name	Author	Scale	License	Dataset Size	Dataset	Pretrained Model
1xESRGAN	victorca25	1	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	4xPSNR
2xESRGAN	victorca25	2	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	4xPSNR
4xESRGAN	victorca25	4	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	4xPSNR
8xESRGAN	victorca25	8	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	4xPSNR
16xESRGAN	victorca25	16	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	4xPSNR
1xPSNR	BlueAmulet	1	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	none
2xPSNR	BlueAmulet	2	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	none
4xPSNR	BlueAmulet	4	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	none
8xPSNR	BlueAmulet	8	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	none
16xPSNR	BlueAmulet	16	Apache License 2.0	65k	Combination of DIV2K, Flickr2K and GOPRO	none

https://drive.google.com/drive/folders/1ldwajXL50uC7PCS63B4Wato6Dnk-svNL

These models were transformed from the original RRDB_ESRGAN_x4.pth model into the other scales, in order to be used as pretrained models for new models in those scales.

More information can be found here.

ESRGAN ("new Architecture") Models

Models that use the "new" ESRGAN architecture. It has no advantages over the old architecture, but breaks compatibility with old arch models and scales other than 1 (if you use the official ESRGAN repo). In the future victorc plans to make his fork compatible with both, providing the option to convert between them. They can be used either with the official ESRGAN repository and BasicSR (not victorc's fork).

If you want to train your own model, please use the "old" architecture instead. There really are no disadvantages to it.

PPON Models

Upscaling

Pixel Art

Name	Author	Scale	License	Purpose	Iterations (Phase 1; 2; 3)	Batch Size (Phase 1; 2; 3)	HR Size (Phase 1; 2; 3)	Dataset Size (Phase 1; 2; 3)	Dataset (Phase 1; 2; 3)	Pretrained Model
Pixie	victorca25	4		Pixel Art / some Cartoons	80K(?; ?; ?)	8; 8; 8	192; 192; 192	?; ?; ?	Custom(Drawings); Custom(Drawings); Custom(Drawings)	PPON
xBRZ+	victorca25	4		Pixel Art	60K(?; ?; ?)	8; 8; 8	128; 128; 128	?; ?; ?	Custom (xBRZ images); Custom (xBRZ images); Custom (Drawings)	Pixie

Pretrained models for different scales

Name	Author	Scale	License	Purpose	Iterations	Batch Size	HR Size	Epoch	Dataset Size	Dataset "
PPON	Zheng Hui (惠政)	4	?	Pretrained model	?	?	?	?	?	?	?

Licenses Links

• GNU GPLv3:
  • If you modify, interpolate or use the model as a pretrained model for your own model and share results of your resulting model, it will have to be under the same license.
  • You have to state that you used the model and its author for your results.
  • You have to state any changes you made to the model.
  • There are other points, but those are the main ones.

---

Testing captcha. Again. Testing again with basic account.