I. Introduction
Thank you for your love and support on my previous articles, I feel energized in a supportive community as such. One interesting thing I noticed — based on the polls from my previous article (35 must-have extensions) - is that there's still a high demand for Stable Diffusion image upscaling tools and extensions. This definitely deserves its own blog. So, I did some research across various sites and found over 160 upscaling models. In this blog, I’ve compiled their information, including their purposes and download links, along with step-by-step instructions for your convenience.
FYI:
A. Previous post result:
Extensions for precise controls, editing and fine-tuning 16.01%
Extensions for image upscaling with details 14.11%
Extensions to create AI animations and videos 12.22%
SDXL compatible tools for A1111 or ComfyUI 10.33%
Extensions for Image generation 15.32%
B. I gathered the information primarily from 2 sources:
a. OpenModelDB, a community wiki that lists various custom models trained on different art styles and subjects.
b. Upscale Wiki, which features models officially released by their original developers, such as ESRGAN.
C. To compare different models interactively, you can check this page
II. 3 Universal Upscalers (Instant Access)
The below 3 models are developed by our team. We trained them on a diverse range of image types - including realistic photos, anime, AI-generated art, and illustrations – with the goal of creating upscalers capable of handling virtually any input while preserving ultra-fine details.
More-detail GAN
Scale: 1X,2X, 4X, 8K
Purpose: upscale images require fine attention to detail, like hair, fur, feathers, and architectural textures.
Smooth Diff
Scale: 1X,2X, 4X, 8K
Purpose: upscale simpler images such as 2D animations, cartoons, glass surfaces, smooth textures, and line drawings. It's also suitable for portraits.
Real-Photo
Scale: 1X, 2X
Purpose: upscale photorealistic images, restore old photos, and improve clarity in high-resolution photos with noise or imperfections.
Download the Models
We've integrated these 3 models into our Aiarty Image Enhancer, eliminating the need for manual installation like with Stable Diffusion.
Download Aiarty for direct access to all three models >
Use the Models
Step 1. Import source images. (you can do this in bulk). Wait for the tool to inference the image and fix the flaws.
Step 2. Choose an upscale model. We choose More Details since the original image contains intricate details in the face texture and hair.
Step 3. Select the target resolution and the tool starts to inference your image and upscaling.
Step 4. Inspect the details in the side-by-side preview window to check if it’s what you need. You can then hit RUN to save the upscaled image locally or experiment with other models and scaling options.
III. 160+ Upscalers for Stable Diffusion
This collection includes some fresh models released just last month, as well as some older ones from 2-3 years ago that are still actively discussed by users. You can use these upscalers not only with Stable Diffusion but also with local programs like chaiNNer, Python scripts, and even some convenient online demo sites. Just be aware they need to be manually installed into your workflow.
40+ Upscalers for Realistic Photos
4x_4xNomos2_hq_dat2
Purpose: upscale photography while fixing realistic noise, lens blur, and other artifacts caused by jpg and webp re-compression.
Scale: 4X
Date: 2024-08-29
Dataset: nomos-v2
Architecture: DAT
Download: https://github.com/Phhofm/models/releases/download/4xNomos2_hq_dat2/4xNomos2_hq_dat2.pth
4x_Nomos2_hq_mosr
Purpose: enhance non-degraded input images with high-quality results
Scale: 4X
Date: 2024-08-25
Dataset: Nomos built by musl
Architecture: MoSR
Download: https://github.com/Phhofm/models/releases/download/4xNomos2_hq_mosr/4xNomos2_hq_mosr_fp32.onnx
4x_Nomos2_realplksr_dysample
Purpose: enhance and upscale low-resolution images while preserving depth of field and mitigating quality loss caused by JPEG compression down to 70%
Scale:4X
Date: 2024-06-30
Dataset: nomos-v2
Architecture: RealPLKSR
4x_NomosWebPhoto_esrgan
Purpose: upscale photography while fixing realistic noise, lens blur, and other artifacts caused by jpg and webp re-compression.
Scale:4X
Date: 2024-06-16
Dataset: Nomos built by musl
Architecture: ESRGAN
Download: https://github.com/Phhofm/models/releases/download/4xNomosWebPhoto_esrgan/4xNomosWebPhoto_esrgan.pth
4x_NomosWebPhoto_atd
Purpose: upscale images with realistic noise, lens blur, and compression artifacts
Scale: 4X
Date: 2024-06-07
Dataset: Nomos-v2
Architecture: ATD
4x_NomosWebPhoto_RealPLKSR
Purpose: enhance and refine photography by simulating realistic noise, lens blur, and compression artifacts from JPG and WebP formats
Scale: 4X
Date: 2024-05-28
Dataset: Nomos-v2
Architecture: RealPLKSR
Download:
https://github.com/Phhofm/models/releases/download/4xNomosWebPhoto_RealPLKSR/4xNomosWebPhoto_RealPLKSR.pth
4x_RealWebPhoto_v4_drct-l
Purpose: upscale realistic images
Scale:4X
Date: 2024-05-02
Dataset: N4xRealWebPhoto_v4
Architecture: DRCT
Download:
4x_RealWebPhoto_v3_atd
Purpose: remove screentone
Scale:4X
Date: 2024-03-22
Dataset: 4xRealWebPhoto_v3
Architecture: ATD
Download: https://github.com/Phhofm/models/releases/download/4xRealWebPhoto_v3_atd/4xRealWebPhoto_v3_atd.pth
4x_Nomos8k_atd_jpg
Purpose: upscale photos while fixing jpg compression (keeping noises)
Scale: 4X
Date: 2024-03-22
Dataset: nomos8k
Architecture: ATD
Download: https://github.com/Phhofm/models/releases/download/4xNomos8k_atd_jpg/4xNomos8k_atd_jpg.pth
4x_RealWebPhoto_v2_rgt_s
Purpose: enhance and upscale low-resolution images while preserving depth of field and mitigating quality loss caused by JPEG compression down to 70%.
Scale:4X
Date: 2024-03-10
Dataset: 4xRealWebPhoto_v2
Architecture: RGT
4x_SPANkendata
Purpose: upscale undegraded photos.
Scale:4X
Date: 2024-01-29
Dataset: developer’s custom dataset, made from an amalgamation of other datasets, as well as images he sourced himself.
Architecture: SPAN
4x_ClearRealityV1
Purpose: upscale realistic photos
Scale:4X
Date: 2023-12-12
Dataset: developers own UltraSharpV2 dataset, my 8k dataset (v2), Nomos8k, and FaceUp
Architecture: SPAN
Download: https://mega.nz/folder/Xc4wnC7T#yUS5-9-AbRxLhpdPW_8f2w
4x_Nomos8kHAT-L_otf
Purpose: upscale realistic photos
Scale: 4X
Date: 2023-11-01
Dataset: Nomos8k
Architecture: HAT
Download:
https://drive.google.com/uc?export=download&confirm=1&id=1_7cmypb7d-GGgGwmWmPg8ugJjOozXrFU
4x_FFHQDAT
Purpose: upscale facial images while handling JPEG compression, blur, and resizing, preserving skin detail
Scale:4X
Date: 2023-08-25
Dataset: FFHQ
Architecture: DAT
Download: https://drive.google.com/uc?export=download&confirm=1&id=1N3ysO2IWkouzy4aFONLgYUjaUMrLz8AB
4x_Nomos8k_span_otf
Purpose: upscale photos
Scale:4X
Date: 2023-08-13
Architecture: SPAN
Download:
Nomos8k_span_otf_weak
https://drive.google.com/uc?export=download&confirm=1&id=1KkV23QH3oBAtl88HBKMq88cdWEv0J2hV
Nomos8k_span_otf_medium
https://drive.google.com/uc?export=download&confirm=1&id=1PyugI5imMP__vLrYYJKvUC_OUO6RcFYG
Nomos8k_span_otf_strong
https://drive.google.com/uc?export=download&confirm=1&id=1K_OUt9lwvDXn280OTfURm0tD3E0lVZ5b
4x_Nomos8kDAT
Purpose: deliver realistic super-resolution images while effectively managing JPEG compression, blur, and resizing
Scale:4X
Date: 2023-08-13
Dataset: Nomos8k_sfw
Architecture: DAT
Download: https://drive.google.com/uc?export=download&confirm=1&id=1PyugI5imMP__vLrYYJKvUC_OUO6RcFYG
4x_RealisticRescaler
Purpose: upscale realistic low-res textures that are compressed by either JPEG or BC1
Scale:4X
Date: 2023-01-01
Dataset: DIV2K, Flick2R, AmbientCG, Poly Haven, nomos2k
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1RcqRuygqR5AwMLCT0olLIqLobs4mBkIi
4x_4xLSDIRPlus
Purpose: upscale small photos with compression, noise and slight blur to 4x their size. Extending my last 4xLSDIRCompact model to Real-ESRGAN, meaning trained on synthetic data instead to handle more kinds of degradations, it should be able to handle compression, noise, and slight blur.
Scale:4X
Date: 2023-07-06
Dataset: LSDIR (84,991 images / 165 GB)
Architecture: ESRGAN
Available versions: 4xLSDIRPlus, 4xLSDIRPlus F, 4xLSDIRPlus C
Download: https://github.com/Phhofm/models/tree/main/4xLSDIRplus
4x_lsdirdat
Purpose: upscale photo by 4x of the original size while fixing blurriness and other issues.
Scale: 4X
Date: 2023-07-06
Dataset: LSDIR
Architecture: DAT
Download: https://drive.google.com/uc?export=download&confirm=1&id=1njK_72D2bXV5-t0GYqOItpm_jrPh2ylw
4x_Nomos8kSCSRFormer
Purpose: produce realistic super-resolution images while handling on-the-fly JPEG compression and blur
Scale: 4X
Date: 2023-06-26
Dataset: nomos8k_sfw
Architecture: SRFormer
Download: https://drive.google.com/uc?export=download&confirm=1&id=1VpMity5vqrEN7YFaPwawhVTdWYdJK8DG
StableSR
Purpose: an upscaling script designed to preserve intricate details while maintaining the facial identity of characters, all while requiring less VRAM.
Scale: up to 16X
Date: 2023-05-27
Dataset: DIV2K
Note: choose v2-1_512-ema-pruned as the checkpoint when you us it in Stabel Diffusion
Download: https://github.com/IceClear/StableSR/tree/main
4x_Nomos8kSC
Purpose: upscale realistic photos
Scale: 4X
Date: 2023-05-10
Dataset: Nomos8k_sfw
Architecture: ESRGAN
Download: https://github.com/Phhofm/models/raw/main/4xNomos8kSC/4xNomos8kSC.pth
4x_LSDIRCompact
Purpose: upscale compressed photos to x4 their size. Able to handle JPG compression (30-100).
Scale:4X
Date: 2023-04-11
Dataset: LSDIR
Architecture: ESRGAN
Available versions: 4xLSDIRCompactC3, 4xLSDIRCompactN3, 4xLSDIRCompactR3
Download: https://github.com/Phhofm/models/tree/main/4xLSDIRCompact/Version3
4x_Rybu
Purpose: general realistic super-resolution.
Scale: 4X
Date: 2023-04-01
Dataset: Nomos8k
Architecture: Compact
Download: https://objectstorage.us-phoenix-1.oraclecloud.com/n/ax6ygfvpvzka/b/open-modeldb-files/o/4x-Rybu.pth
4x Valar
Purpose: upscale realistic images
Scale: 4X
Date: 2021-09-16
Dataset: hand selected 1471 Raw images from the Adobe-MIT 5k dataset, processed using Rawtherapee
Architecture: ESRGAN
Download: https://mega.nz/file/STI3kIKJ#mr3PqAu35SI9yGHWUdCJsXm2kai1R1SrsYsBafyH-Vo
4x_Fabric
Purpose: upscale fabric or cloth textures (works on cats too!)
Scale: 4X
Date: 2021-9-15
Architecture: ESRGAN
Download: https://mega.nz/folder/6EJjwADL#oDvv9p0z1t9nQvObbvBRBQ
4x_CountryRoads
Purpose: upscale streets with dense foliage in the background. Upscale outdoor scenes.
Scale: 4X
Date: 2021-08-26
Dataset: Streetside photos taken on Samsung Galaxy A21 camera with all enhancements disabled using GCam and FreeDCam, downscaled to 1024
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1CQf6nLYQQiWMaapDUpD0S08de4uUGzfN
4x_NMKD Superscale
Purpose: upscale realistic images/photos while fixing noise and compression artifacts
Scale: 4X
Date: 2020-07-22
Architecture: ESRGAN
Download: https://icedrive.net/1/43GNBihZyi
4x_NickelbackFS
Purpose: upscale most pictures/photos (granted they are clean enough) without destroying as much detail as Box and basic ESRGAN.
Scale: 4X
Date: 2020-07-10
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1GL0OjGPsMwSmCr_FmkqtRjZu2IM0s019It
4x_NickelbackFS
Purpose: It can upscale most pictures/photos (granted they are clean enough) without destroying too much noises
Scale:4X
Date: 2020-07-10
Dataset: Newspapers
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1GL0OjGPsMwSmCr_FmkqtRjZu2IM0s019
4x_RealSR DF2K
Purpose: upscale corrupted images with processing noise
Scale: 4X
Date: 2020-05-26
Dataset: DF2K
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1pWGfSw-UxOkrtbh14GeLQgYnMLdLguOF
4x_RealSR DF2K (JPEG)
Purpose: upscale compressed jpeg image
Scale: 2X
Date: 2020-05-26
Dataset: DF2K
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1w8QbCLM6g-MMVlIhRERtSXrP-Dh7cPhm
4x_RealSR DPED
Purpose: upscale real images taken by cell phone camera.
Scale:4X
Date: 2020-05-26
Dataset: DF2K
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1zZIuQSepFlupV103AatoP-JSJpwJFS19
4x_ Nickelback
Purpose: upscale most pictures/photos (granted they are clean enough) without destroying as much detail as the aforementioned models. It generates less moiré like patterns and keeps details without oversharpening or blurring the image too much.
Scale:4X
Date: 2020-02-12
Dataset: Newspapers
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1d89zvzC6BKf5EKapBtc4MH2DX5GClKzl
4x_NickelbackFront
Purpose: upscale coins
Scale: 4X
Date: 2020-02-12
Dataset: coin images
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1zC97RLUg5ejqb5-4Dj5q-SzdLPLiMozH
4x_FSDedither
Purpose: upscale photos/realistic images, but worth trying on other images that have reduced colors and dithering along with fine details
Scale: 4X
Date: 2020-01-29
Dataset: Div2K+Flickr2K, linearly downscaled and dithered (half Floyd-Steinberg, half ordered) using GIMP.
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1H4KQyhcknOoExjvDdsoxAgTBMO7JuJ3w
4x_SmoothRealism
Purpose: upscale pixel art, rocky/grainy textures while adding detail
Scale: 4X
Date: 2020-03-04
Dataset: LR: Default minecraft. HR: R3DCraft Smooth Realism Texture Pack (64x resolution)
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1Uc9RUc2YpZKpPoGQeGxNUb3Ro-U720cH
4x_LADDIER1
Purpose: remove noise, grain, box blur, lens blur and gaussian blur and increase overall image quality
Scale: 4X
Date: 2019-11-13
Dataset: Mostly images of Nirvana some of other images
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1GoUY7t5kE0Ubi-NDvlMB8lUs-_u_f6tS
4x_ESRGAN
Purpose: expertly upscales low-resolution images, achieving impressive real-world image restoration, including the removal of JPEG compression artifacts
Scale:4X
Date: 2019-7-16
Architecture: ESRGAN
4x_Realistic Misc
Purpose: upscale realistic images
Scale: 4X
Date: 2019-05-03
Dataset: various pictures shot by the developer, including bricks, stone, dirt, grass, plants, wood, bark, metal and a few others.
Architecture: ESRGAN
Download: https://e.pcloud.link/publink/show?code=XZ6msZu8BOExSFH4SxixhuxoyrmbRYGQQk
SwinIR Upscaler
Purpose: restore high-quality images from low-quality inputs with noises and compression artifacts.
Scale: 2x to 8x (varies by specific upscaler)
Dataset: https://github.com/JingyunLiang/SwinIR/tree/main/testsets
Architecture: ESRGAN
Available models include:
SwinIR-M (DF2K-s64w8): x2, x3, x4, x8
SwinIR-M (DIV2K-s48w8): x2, x3, x4, x8
SwinIR-L-x4-GAN (realSR-BSRGAN-DFOWMFC-s64w8)
SwinIR-M-GAN (realSR-BSRGAN-DFO-s64w8): x2, x4
Downloads:
Full collection download: https://civitai.com/api/download/models/168140?type=Archive&format=Other
Individual models: https://github.com/JingyunLiang/SwinIR/releases/
70+ Upscalers for Anime/Cartoon/Manga/Pixel Art/Game
4x_DWTP DS dat2 v3 2
Purpose: upscale photos downloaded from the web
Scale: 4X
Date: 2024-02-26
Architecture: DAT
Download: https://drive.google.com/uc?export=download&confirm=1&id=1r8uTdQ1kaMnKAZr6XSStB10JPTzOOFov
4x_eula digimanga MiA 65k
Purpose: upscale grayscale pencil/sketch style art with paper textures (made in abyss specific)
Scale: 4X
Date: 2024-01-23
Dataset: 600dpi scans of from Made in Abyss manga
Architecture: ESRGAN
Download:
4x_Fatality Comix
Purpose: upscale halftone images.
Scale:4X
Date: 2023-10-25
Architecture: ESRGAN
Download: https://drive.proton.me/urls/YPSHVH1N1G#pOejdrktCIaD
2x_Garfield1_308k
Purpose: upscale DVDs while fixing rainbowing, dot-crawl, halos, general noise, and MPEG-2 artifacts.
Scale: 2X
Date: 2023-05-23
Dataset: upscaled DVD frames using filtered and raw DVD source with hundreds of manual fixes
Architecture: ESRGAN
Download: https://mega.nz/file/nWIXFD7K#EipZnSIKc6PLf6UWNkG88BybAHX_bCs5qwvu13wiW90
2x_Gen5 Alpha
Purpose: Upscale Pixel Art with transparency/alpha channel
Scale: 2X
Date: 2023-05-23
Dataset: 5 Pokemon sprites
Architecture: ESRGAN
Download: https://mega.nz/file/nWIXFD7K#EipZnSIKc6PLf6UWNkG88BybAHX_bCs5qwvu13wiW90
4x_HFA2k
An anime images 4x upscaling model based on musl's HFA2k dataset
Scale: 4X
Date: 2023-05-07
Dataset: Manga109 copied a bunch of times with different dithering methods, and both linear and nearest-neighbor scaling.
Architecture: ESRGAN
Examples: https://imgsli.com/MTc2NDgx (PS Example input images are small, each around 500x280 px)
Download: https://github.com/Phhofm/models/raw/main/4xHFA2k/4xHFA2k.pth
2x_90s Cartoon v1
Purpose: upscale 90s and early 2000 American cartoons
Scale: 2X
Date: 2023-03-10
Data set: HD rescans of some 90s cartoons and some cels, i also used 4xfilmfream V1 as teacher for some of the images in the dataset
Architecture: Compact
Download: https://drive.google.com/uc?export=download&confirm=1&id=1V_6JfE7wu1uCaOPpMq2E4wpSGxB-81XY
2x_GT-v2-evA-01
Purpose: Upscale Dragon ball GT
Scale:2X
Date: 2023-03-01
Dataset: German blu-ray of detective Conan and other shows and some cels
Architecture: Compact
Download: https://huggingface.co/gemasai/2xGT-v2-evA-01/blob/main/2xGT-v2-evA-01.pth
2x_AnimeJaNai v2 Compact
Purpose: Upscale animes to 4K while fixing scaling artifacts and blur
Scale: 2X
Date: 2023-03-01
Dataset: German blu-ray of detective Conan and other shows and some cels
Architecture: Compact
Download: https://huggingface.co/gemasai/2xGT-v2-evA-01/blob/main/2xGT-v2-evA-01.pth
2x_Futsuu Anime
Purpose: Upscale animes
Scale: 2X
Date: 2023-01-08
Dataset: frames from a wide variety of animated sources
Architecture: Compact
4x_HDCube3
Purpose: Gamecube and Wii textures.
Scale: 4X
Size:63.9MB
Date: 2022-12-23
Architecture: ESRGAN
Download: https://github.com/Venomalia/HDcube/tree/main/v3#hdcube-3
4x_Escale
Purpose: upscale anime/visual novel art
Scale: 4X
Date: 2022-12-17
Dataset: various eroge dumps
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=15VfEwR61Y1Je8EyPMWl40wW01S6bJuPJ
4x_eula_anifilm v1
Purpose: upscale cel animation
Scale: 4X
Date: 2022-12-14
Dataset: Dragon Ball Level sets
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=15VfEwR61Y1Je8EyPMWl40wW01S6bJuPJ
4x_HDCube3
Purpose: it can be used for all image formats supported by Gamecube and Wii hardware and can remove its typical artifacts like CMPR Block Compression (DXT1 algorithm, also known as BC1), color palette errors, color reduction up to 8bit color depth and 1bit alpha depth.
Scale: 4X
Date: 2022-12-13
Dataset: Artifact free textures and photos (stone, wood, metal, clothes, plants, characters, FX)
Architecture: ESRGAN
Download: https://github.com/Venomalia/HDcube/tree/main/v3#hdcube-3
2x_Bubble AnimeScale Compact v1
Purpose: Upscale animes without losing lot of details
Scale: 2X
Date: 2022-11-13
Dataset: SFW and NSFW anime frames that were sourced from Blu Ray discs and high bitrate Web releases. These were then degraded with various different methods
Architecture: Compact
2x_Bubble AnimeScale SwinIR Small v1
Purpose: Upscale anime frames faithfully without major contrast shifting
Scale: 2X
Date: 2022-11-13
Dataset: SFW and NSFW anime frames that were sourced from Blu Ray discs. These were then degraded with various different methods.
Architecture: SwinIR
2x_DigiGradients Lite
Purpose: Upscale the TMNT 2003 DVDs.
Scale: 2X
Date: 2022-08-24
Dataset: Images from The Batman (series) HD WEB-DL
Architecture: ESRGAN
4x_SmolFace
Purpose: upscale small sprite faces. Does not blend, so avoid using on painted/photo portraits unless you were trying to retain more of the outlines somehow.
Scale:4X
Date: 2022-08-19
Dataset: FatalityFaces + danbooru side faces
Architecture: ESRGAN
Download: https://1drv.ms/u/s!Aip-EMByJHY29BRdcYSnTpgx0LWd?e=I6x5R7
2x_anifilm-compact
Purpose: Upscale anime images at a faster speed (might soft some images).
Scale:2x
Date: 2022-08-02
Dataset: Dragon Ball Z Movies
Architecture: Compact
2x_DigitalFlim SuperUltraCompact
Purpose: Quickly upscale images while removing noises
Scale: 2X
Date: 2022-05-14
Architecture: ESRGAN
4x_GameAI 1.0
Purpose: mainly handle PS2 compression and a mixture of Realistic and cartoonish textures. It's not meant to be used for very low resolution textures such as item icons
Scale: 4X
Date: 2022-02-27
Dataset: Textures from Skyrim, The Witcher 3, Resident Evil 4, Final Fantasy XV and Tales of Vesperia. 21.9k to 70k textures, up to 1024x1024 per texture.
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=15KhApV05F8nEqTa5ZjLE821rhe1k8VOg
4x_GameAI 2.0
Purpose: mainly handle PS2 compression and a mixture of Realistic and cartoonish textures. It's not meant to be used for very low resolution textures such as item icons
Scale:4X
Date: 2022-02-27
Dataset: Textures from Skyrim, The Witcher 3, Resident Evil 4, Final Fantasy XV and Tales of Vesperia. 21.9k to 70k textures, up to 1024x1024 per texture.
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=15KhApV05F8nEqTa5ZjLE821rhe1k8VOg
4x-AnimeSharp (aka 4x-TextSharpV1)
Purpose: upscale anime. It also upscales text, but it's far better with anime content.
Scale:4X
Date: 2021-12-28
Architecture: ESRGAN
Download: https://mega.nz/file/CFZGHb4A#yo2NaVANd9-gk9n4lxA9zkzKPqVukZYpzendRVCinDw
4x_BooruGan 600k
Purpose: mainly upscale anime artworks.
Scale:4X
Date: 2021-11-26
Dataset: high-res artworks from gelbooru
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1DRyqqy24OU6G_mwsGjy6Zaxan_Fy1LNU
4x_BooruGan 650k
Purpose: mainly upscale anime artworks. If you have issues with chroma then try the 600k iterations release.
Scale:4X
Date: 2021-11-26
Dataset: high-res artworks from gelbooru
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1psKp2DLWscPmDuzSTC8XqXL0LWnWuTs
4x_VolArt
Purpose: upscale artwork for the game Volfoss (2001) and it peaked in quality very quickly. The NR model removes most noise, but may remove the transparent portions. Use the main model in most cases.
Scale:4X
Date: 2021-09-18
Dataset: Custom Dataset consisting of lossless 4k frames from Metal Arms: Glitch in the System, Just Cause 3, Dirt 3, Forza Horizon 3, Sleeping Dogs, and self-edited photos from SignatureEdits
Architecture: ESRGAN
Download: https://mega.nz/folder/HNh0hLzD#vUuFhp2ZAeNi_5shZfCC0g
2x_UniScale CartoonRestore Lite
Purpose: This model has very strong compression removal and line restructuring, allowing it to restore any heavily compressed drawings, animation, cartoons, or anime. It also works on games as well as DDS compression, rendering frames quickly and making it viable for restoring videos.
Scale: 2X
Date: 2021-08-31
Dataset: Custom dataset consisting of lossless 4k frames from Metal Arms: Glitch in the System, Just Cause 3, Dirt 3, Forza Horizon 3, Sleeping Dogs, ATLA DVD, and self-edited photos from SignatureEdits.
Architecture: ESRGAN
2x_UniScale CartoonRestore Lite
Purpose: restore heavily compressed drawings, animation, cartoons, or anime. Also works on games as well as DDS compression. It renders frames very quickly and is very viable for restoring videos.
Scale: 2X
Date: 2021-08-31
Dataset: Custom Dataset consisting of lossless 4k frames from Metal Arms: Glitch in the System, Just Cause 3, Dirt 3, Forza Horizon 3, Sleeping Dogs, ATLA DVD, and self-edited photos from SignatureEdits
Architecture: ESRGAN
4x_DigitalFake 2.1
Purpose: upscale cartoon and anime
Scale:4X
Date: 2021-08-26
Dataset: DIV2K and spongebob frames, downscaled with box and upscaled with Digitalframes 2.1
Architecture: ESRGAN
Download:
https://u.pcloud.link/publink/show?code=XZQlqjXZPNgmLRJl5HLLG0M3VRmzmzmHr5hk
GFPGAN
Purpose: upscale photos while addressing the common issue of facial image degradation in low-resolution images, particularly in the context of image restoration tasks.
Scale:1X -8X
Date (first release): 2021-06-15
Dataset: FFHQ (Flickr-Faces-HQ), CelebA
Architecture: GFPGAN
Download: https://github.com/TencentARC/GFPGAN/releases
NXbrz
Purpose: basic pixel art upscaling, for people who want a more simpler style and lightweight pixel art upscaling model
Scale: 4X
Date: 2021-06-09
Dataset: Images upscaled with the Xbrz scaling filter and images upscaled with the ScaleNX filter, combined together to create a mix of both models.
Architecture: ESRGAN
4x_Loyaldk Kororo
Purpose: upscale Anime while keeping as much of the original detail as possible. Isn't among the sharpest models and if the anime is somewhat blurry it will retain that detail but add more pixels to give a smoother edge. Doesn't change the color very much and may make some scene's very slightly darker.
Scale: 4X
Date: 2021-06-07
Dataset: Saga of Tanya The Evil, Gravity Falls, Lucky Star, One Punch Man, Redline, Soul Eater, Spirited Away, Your Lie in April and Your Name. Converted Episodes and Movies to 4:4:4 Chroma Subsampling using custom 1x model included. LR's Include H264 Compression, H265 Compression, 1x_DotCrawl_125000_G, 1x_DeteriorationFrames V1_102000_G, Gaussian Blur, Poisson Noise, Uniform Noise, JPEG Compression, 4:2:0 Chroma Subsampling, Dithering.
Architecture: ESRGAN
2x_MangaScaleV3
Purpose: Upscale manga
Scale: 2X
Date: 2021-05-24
Dataset: Self curated custom manga dataset
Architecture: Compact
Download: https://mega.nz/file/nR51EYYC#RcAKkxb8hAzvZAj4TLxnAukd9gbrYuUeyjUp2l1ffb8
4x_UltraFArt v3
Purpose: upscale art and illustrations with larger shaped features.
Scale: 4X
Date: 2021-05-12
Dataset: ArtStation + Fatality Faces tiles.
Architecture: ESRGAN
Available versions: UltraFArt v3 Photo, UltraFArt v3 Fine, UltraFArt v3 Smooth
Download: https://1drv.ms/u/s!Aip-EMByJHY27UIFr1mk87zFytap?e=SSxKaV
4x_Loyaldk SuperPony V2.0
Purpose: Upscale MLP episodes. Full version. Able to handle compression better compared to V1.0 and no longer creates halo's/rainbows. Good for Vector 2D art however converts detail to blobs.
Scale:4X
Date: 2021-04-30
Dataset: All MLP: FiM Episodes using scene detection frame grabbing as well as selected Prores4444 content https://shimmermare.com/4k-pony-project. Converted Episodes and Movies to 4:4:4 Chroma Subsampling using custom 1x model. LR's Include H264 Compression, H265 Compression, 1x_DotCrawl_125000_G, 1x_DeteriorationFrames V1_102000_G, Gaussian Blur, LitePony Chroma Bleed, Uniform Noise, JPEG Compression, 4:2:0 Chroma Subsampling.
Architecture: ESRGAN
Download: https://mega.nz/file/BMhUCbJZ#kYz0wvc26RW0n0f8AUXekW_ZCuRamKmh2-iWf1eefn0
4x_Loyaldk MediumPony V2.0
Purpose: upscale MLP episodes. Full version. Able to handle compression better compared to V1.0 and no longer creates halo's/rainbows. Good for Vector 2D art however converts detail to blobs.
Scale: 4X
Date: 2021-04-30
Dataset: All MLP: FiM Episodes using scene detection frame grabbing as well as selected Prores4444 content https://shimmermare.com/4k-pony-project. Converted Episodes and Movies to 4:4:4 Chroma Subsampling using custom 1x model. LR's Include H264 Compression, H265 Compression, 1x_DotCrawl_125000_G, 1x_DeteriorationFrames V1_102000_G, Gaussian Blur, LitePony Chroma Bleed, Uniform Noise, JPEG Compression, 4:2:0 Chroma Subsampling.
Architecture: ESRGAN
4x_Loyaldk LitePony V2.0
Purpose: Good for Vector 2D art however converts detail to blobs. Quality seems weak with this one.
Scale:4X
Date: 2021-04-12
Dataset: All MLP: FiM Episodes using scene detection frame grabbing as well as selected Prores4444 content https://shimmermare.com/4k-pony-project. Converted Episodes and Movies to 4:4:4 Chroma Subsampling using custom 1x model. LR's Include H264 Compression, H265 Compression, 1x_DotCrawl_125000_G, 1x_DeteriorationFrames V1_102000_G, Gaussian Blur, LitePony Chroma Bleed, Uniform Noise, JPEG Compression, 4:2:0 Chroma Subsampling.
Architecture: ESRGAN
4x_animerd v1
Purpose: upscale animes
Scale: 4X
Date: 2021-04-01
Dataset: Realistic images
Architecture: ESRGAN
2x_VHS upscale and denoise Film
Purpose: Upscale VHS captures of Film material, but may work on VHS recorded native SD-TV material as well. Also useable for cleaned up source material
Scale: 2X
Date: 2021-03-28
Dataset: For LR, two-decade old VHS recording of a movie from TV, captured with enabled TBC on the VCR. For HR, Blu-ray rip of that movie.
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1LFiaoiPDUoAIfNCTDBm2GzJnWvgjxft9
8x_NMKD-Sphax-DeDither
Purpose: upscale low-resolution MS Paint drawings, general pixel art, and general dithered pixel art, all kinds of pixel art
Scale: 8X
Date: 2021-03-24
Architecture: ESRGAN
2x_KcjpunkAnime 2.0 Lite
Purpose: upscale anime
Scale: 2X
Date: 2021-03-23
Dataset: 1080P BD anime frames
Architecture: ESRGAN
4x_KCJPUNK
Purpose: up-scale digital animation
Scale:4X
Date: 2021-03-23
Dataset: 1080P BD anime frames
Architecture: ESRGAN
Download: https://mega.nz/folder/QsUWWJjb#SIuqcPEzrkpLW0WNSth66g
2x_DigitalFilmV5 Lite
Purpose: upscale Dragon Ball Z DBox DVD's. It keeps some grain, but also does some cleaning, sharpening, and fixing.
Scale: 4X
Date: 2021-02-19
Dataset: DB/DBZ/DBGT Dragon Box DVD frames, upscaled with various models such as DigitalFrames2.0, FilmFrames677k, SharpAnimeV2, and manually edited to only show the best parts of each model. This process is referred to as image stacking.
Architecture: ESRGAN
PocketMonsters-Alpha
Purpose: upscale pixel art with alpha channels that should perform better than any other. It should work well on both cartoon and 3D styled content.
Scale:4X
Date: 2021-02-04
Dataset: Pokemon drawings, vector art, and Pokemon Home 3D render sprites, downscaled to 256x256 for HR and downscaled with nearest neighbor for LR.
Architecture: ESRGAN
Download: https://u.pcloud.link/publink/show?code=kZrIFhXZxnpi72FuwgLz7FIDw8kdrp2suIEk
4x_SGI
Purpose: upscale and ditherier pre-rendered sprites, images and textures made in the 90s.
Scale:4X
Date: 2020-12-17
Dataset: Very hi-res (up to 2500px) vintage cgi images and prerendered promotional renders for games from the 90s to early 2000s, as well as HD stills from movies made with SGI workstations (early pixar movies and shrek)
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1kNSz5f2_krdlsJ700DwH78OEPYkkcjW5
4x_BS_ScreenBooster_SPSR
Purpose: upscale game screenshots (3D Games) by 4 times
Scale:4X
Date: 2020-12-17
Dataset: devoloper’s own screenshots
Architecture: SPSR
Download: https://drive.google.com/uc?export=download&confirm=1&id=1JI8VZSSHpH9u7HzE8MvuNG351Dgn4jwU
4x_FSMangaV2
Purpose: upscale manga-style images with or without dithering - cartoons, maybe pixel art, etc
Scale:4X
Date: 2020-12-14
Dataset: Manga109 copied a bunch of times with different dithering methods, and both linear and nearest-neighbor scaling.
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1uraeNuUPx46HKZ77xUI1PCBBoH-LhikY
2x_BIGOLDIES
Purpose: Upscale old anime. help to denoise and find lines and dehalo
Scale:2X
Date: 2020-12-09
Dataset: LR from several dvd anime and HR from blurry anime
Architecture: ESRGAN
2x_SHARP ANIME V2
Purpose: Upscale animes which have fairly fine lines at the base
Scale: 2X
Date: 2020-12-09
Dataset: LR from several dvd anime and HR from bluray anime
Architecture: ESRGAN
PixelPerfectV4
Purpose: upscale Pixel Art/Sprites
Scale: 4X
Date: 2020-11-16
Dataset: random anime images
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1x583PgpY5O4T2603gJK8kwrGiIc4aFne
2x_pokemodel lite
Purpose: upscale old anime like pokemon
Scale: 2X
Date: 2020-11-01
Dataset: images from pokemon episodes
Architecture: ESRGAN
4x_Struzan
Purpose: upscale airbrush/pencil-based artwork
Scale:4X
Date: 2020-10-12
Dataset: High-res movie poster artwork by Drew Struzan and Bill Eaken
Architecture: ESRGAN
4x_Link
Purpose: Chainmail game textures. Alternatively, it can be used to turn plain images into chainmail.
Scale: 4X
Date: 2020-10-06
Dataset: pexels.com, ambientcg.com, Wikimedia Commons. photos of chainmail from DeviantArt, game textures with chainmail, diffuse maps of chainmail from DAZ items. All items downscaled by factors of 2 to add capacity for LOD texture upscaling and increase generative effect of model.
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=17gKWlbwLiPGi79vqRaIbbMvsPwI5IxWi
4x_MeguUp
Purpose: upscale lossless (uncompressed) anime art.
Scale:4X
Date: 2020-09-16
Dataset: SFW and NSFW anime artworks
Architecture: ESRGAN
4x_NMKD PatchySharp
Purpose: upscale clean images or images with compression artifacts (jpeg quality >75) - Produces very sharp lines/edges due to NN-Filtered HR images.
Date: 2020-09-12
Dataset: DIV2K Clean + JPEG
Architecture: ESRGAN
Download: https://icedrive.net/1/43GNBihZyi
2x_fidelbd pokemodel
Purpose: Upscale old anime that looks blurry
VHS captures of Film material, but may work on VHS recorded native SD-TV material as well. Also useable for cleaned up source material
Scale: 2X
Date: 2020-08-29
Dataset: Made with images from the first episode of pokemon released on blu ray for the HRs and 112.5% upscaled images for the LRs to match 1/2x the HRs
Architecture: ESRGAN
Download: ttps://objectstorage.us-phoenix-1.oraclecloud.com/n/ax6ygfvpvzka/b/open-modeldb-files/o/2x-fidelbd-pokemodel.pth
4x_BSDevianceMIP
Purpose: upscale digital drawings of many different styles.
Scale:4X
Date: 2020-07-25
Dataset: afew drawings from DA
Architecture: ESRGAN
Download: ttps://drive.google.com/uc?export=download&confirm=1&id=1OVqM2D-T5meRBCdFjsGVl0bgQZd8kZe5
4x_Jaypeg90
Purpose: upscale 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.
Scale: 4X
Date: 2020-07-20
Dataset: Flickr2K+Div2K, JPEG compressed with random quality from 85-95 and scaled with either cubic or NN scaling.
Architecture: ESRGAN
Download: ttps://drive.google.com/uc?export=download&confirm=1&id=17k72r01xTkeOzqsaYCN6ZqnAktKqTLyw
8x_HugePaint
Purpose: upscale digital illustrations
Scale: 8X
Date: 2020-06-29
Dataset: a variety of images from ArtStation
Architecture: ESRGAN
Download: https://1drv.ms/u/s!Aip-EMByJHY26wU-wOqApp7Hu7tt?e=4ihnBr
4x_Fatal_Anime
Purpose: upscale animes
Scale:4X
Date: 2020-06-17
Dataset: Anime, Manga
Architecture: ESRGAN
8x_HugePeeps v1
Purpose: upscale painted humans
Scale: 8X
Date: 2020-06-16
Architecture: ESRGAN
Download: https://1drv.ms/u/s!Aip-EMByJHY26nuB0Lo0rdirfGoi?e=XI9Fa2
4x_Fatal Pixels
Scale: 4X
Date: 2020-06-02
Dataset: anime, manga
Architecture: ESRGAN
4x_BigFArt Bang1
Purpose: upscale art/Pixel Art
Scale: 4X
Date: 2020-05-31
Architecture: SPSR
Download: https://1drv.ms/u/s!Aip-EMByJHY23if0ac150Sj2qnbp?e=qLp7SP
4x_BigFArt Base
Purpose: upscale art/Pixel Art
Scale:4X
Date: 2020-05-31
Architecture: SPSR
Download:
https://1drv.ms/u/s!Aip-EMByJHY23if0ac150Sj2qnbp?e=qLp7SP
4x_BigFArt Blend
Purpose: upscale art/Pixel Art
Scale:4X
Date: 2020-05-31
Architecture: SPSR
Download: https://1drv.ms/u/s!Aip-EMByJHY23if0ac150Sj2qnbp?e=qLp7SP
4x_BigFArt Detail
Purpose: upscale art/Pixel Art
Scale:4X
Size:63.8MB
Date: 2020-05-31
Architecture: SPSR
Download: https://1drv.ms/u/s!Aip-EMByJHY23if0ac150Sj2qnbp?e=qLp7SP
4x_BigFArt Fine
Purpose: upscale art/Pixel Art
Scale:4X
Size:63.8MB
Date: 2020-05-31
Architecture: SPSR
Download: https://1drv.ms/u/s!Aip-EMByJHY23if0ac150Sj2qnbp?e=qLp7SP
4x_FSDedither Manga
Purpose: upscale cartoons/pixel art/other non-realistic stuff with dithering
Scale:4X
Date: 2020-04-21
Dataset: Manga109 copied 14 times with different dithering and scaling
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1Vtwt9kgnmju1edibl07WI6Rg7CrAlHIx
4x_BSDeviance
Purpose: upscale digital drawings of many different styles.
Scale:4X
Date: 2020-02-25
Dataset: a few drawings from DA
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1hLhsp3aKurLys4cALQiEkYxyxGWNf1ET
2x_BSWolly
Purpose: Upscale Pixar Movies or Wall-E pictures/frames
Scale:2X
Size:63.7MB
Date: 2020-02-07
Dataset: BD 1080p of Wall-E (FFMPEG extraction at 1FPS with frames of the movie except intro but forgot the credits oops X)
Download: https://drive.google.com/uc?export=download&confirm=1&id=1zlc8BWOcpCbSZpwf7ArMCqpb0Tf9DlZk
4x_Lady0101
Purpose: upscale pixel art/paintings to digital painting style
Scale:4X
Date: 2019-07-25
Dataset: LSDIR
Architecture: ESRGAN
Download: https://1drv.ms/u/s!Aip-EMByJHY20xsc9EvnxVZ50Ja7?e=Loa5Gz
4x_Scalenx
Purpose: scalenx style pixel art upscaler
Scale: 4X
Date: 2019-06-15
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1jE2le5Lsab-AcMUMm0Zr9L-OvFNhnRAG
20+ Upscalers for General Purpose
NomosUni span multijpg ldl
Purpose: fast universal DoF preserving upscaler
Scale: 2X
Date: 2024-01-11
Dataset: nomosuni
Architecture: SPAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1TpqLUsHvqgbnpGr8hOE47BYIH3zqBDpu
Omni-SR
Purpose: used for lightweight image super-resolution, enhancing effective receptive field interactions across both spatial and channel dimensions through omni self-attention and multi-scale interaction schemes.
Scale: 4X,3X,2X
Date: 2023-04-19
Dataset: DIV2K
Architecture: OmniSR
Available version: OmniSR_X2_DIV2K, OmniSR_X2_DF2K, OmniSR_X3_DF2K, OmniSR_X3_DIV2K, OmniSR_X4_DIF2K, OmniSR_X4_DIV2K
Download: https://github.com/Francis0625/Omni-SR
4x_UltraMix
Scale: 4X
Date: 2023-9-19
Architecture: ESRGAN
Available 3 version: UltraMix Balanced, UltraMix Restore, UltraMix Smooth
Download: https://civitai.com/models/147806/ultramix
4x_NomosUniDAT_bokeh_jpg
Purpose: perform high-quality 4x upscaling on a wide range of image types (photographs, anime, text, maps, paintings, etc.) while handling various compression artifacts
Scale: 4X
Date: 2023-09-14
Dataset: nomos_uni
Architecture: DAT
Download: https://drive.google.com/uc?export=download&confirm=1&id=1SBIn0quWcqxq4PAj7UxJ7cqNih-AZwO-
8x_NMKD-Superscale
Purpose: upscale realistic images and digital illustrations
Scale: 8X
Date: 2023-02-04
Architecture: ESRGAN
Download: https://huggingface.co/uwg/upscaler/resolve/main/ESRGAN/8x_NMKD-Superscale_150000_G.pth
4x_UltraSharp
Purpose: restore highly compressed images and work best on JPEG compressions. It generates lots and lots of detail and leaves a nice texture on images.
Scale: 4X
Date: 2021-10-27
Dataset: RAW images shot by myself, SignatureEdits, AdobeMIT-5K, DIV2K, TLOK from brucethemoose, some rock/stone images from ALSA, and many images provided by <@724494344270381097
Architecture: ESRGAN
Download: https://huggingface.co/datasets/BlodyTraveler/4x-UltraSharp/resolve/main/4x-UltraSharp.pth
4x_Remacri
Purpose: upscale images by keeping more details and less smoothing
Scale: 4X
Date: 2021-04-09
Architecture: ESRGAN
Download: https://u.pcloud.link/publink/show?code=kZgSLsXZ0M1fT3kFGfRXg2tNtoUgbSI4kcSy
4x_UniversalUpscalerV2
Purpose: upscale genera images
Date: 2021-04-01
Dataset: Realistic images
Architecture: ESRGAN
Available versions: UniversalUpscalerV2 Neutral, UniversalUpscalerV2 Sharp,
Download: https://huggingface.co/uwg/upscaler/tree/main/ESRGAN/UniversalUpscaler
4x_RealESRGAN
Purpose: upscale general images and videos
Scale: 2X/4X
Architecture: ESRGAN
Available versions:
RealESRGAN AnimeVideo v3 (2022-08-24): Improved results and faster inference speed.
RealESRGANv2 AnimeVideo xs-x4 (2021-12-12): Anime video model with XS size.
RealESRGAN_x4plus_anime_6B (2021-09-01): Much smaller model size.
RealESRGAN_x4plus (2021-08-26)
RealESRGAN_x2plus (2021-08-26)
And more
Download: https://github.com/xinntao/Real-ESRGAN/releases
4x_UniScale
Purpose: upscale images with options for sharper or softer outputs
Date: 2021-8-23
Dataset: custom Dataset consisting of lossless 4k frames from Metal Arms: Glitch in the System, Just Cause 3, Dirt 3, Forza Horizon 3, Sleeping Dogs, and self-edited photos from SignatureEdits
Architecture: ESRGAN
Available versions: Uniscale Balanced, Uniscale Strong, Uniscale V2 Soft, Uniscale V2 Moderate, Uniscale V2 Sharp, Uniscale NR Balanced, Uniscale NR Strong, Uniscale Interp.
Download: https://huggingface.co/JimmyYHung/ESRGAN/tree/main
4x_NMKD Siax ("CX")
Purpose: upscale clean and slightly compressed images (JPEG quality 75 or better)
Scale: 4X
Date: 2020-11-06
Dataset: DIV2K Train+Valid
Architecture: ESRGAN
Download: https://icedrive.net/1/43GNBihZyi
4x_Lollypop
Purpose: enhance prerendered images across various art styles, including realistic faces, manga, pixel art, and de-dithering
Scale: 4X
Date: 2020-09-11
Dataset: prerendered manga images
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=10h8YXKKOQ61ANnwLjjHqXJdn4SbBuUku
4x_FSDedither Riven
Purpose: upscale images from Riven, but should be better in general, particularly on ordered dithering
Scale:4X
Date: 2020-07-11
Dataset: Photos - DIV2K+Flickr2K with different types of dithering applied
Architecture: ESRGAN
4x_FSDedither Riven
Purpose: Fine-tuned 4xFSDedither to upscale images from the game Riven, but should be better in general, particularly on ordered dithering.
Scale:4X
Date: 2020-07-11
Dataset: Photos - DIV2K+Flickr2K with different types of dithering applied
Architecture: ESRGAN
4x_FSDedither
Purpose: upscale photos/realistic images, but worth trying on other images that have reduced colors and dithering along with fine details.
Scale: 4X
Date: 2020-01-9
Dataset: Div2K+Flickr2K, linearly downscaled and dithered (half Floyd-Steinberg, half ordered) using GIMP
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1H4KQyhcknOoExjvDdsoxAgTBMO7JuJ3w
4x_Trixie
Purpose: generate and enhance character textures, primarily for Star Wars games, focusing on heroes, rebels, Sith, Imperials, and key aliens, while also offering general-purpose face texture upscaling
Scale: 4X
Date: 2019-05-09
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=1HIBRKFs7s-XhpN1p7rwAWcMTHwFRqn3n
10+ Upscalers for Face
4x_ArtFaces_realplksr_dysample
Purpose: a dysample RealPLKSR 4x upscaling model for art / painted faces
Scale: 4X
Date: 2024-08-08
Dataset: ArtFaces
Architecture: RealPLKSR_dysample
4x_FaceUp
Purpose: enhance facial images with superior quality
Scale: 4X
Date: 2023-09-02
Dataset: FaceUp dataset
Architecture: DAT
Download available versions:
4xFaceUpDAT (for good quality input)
https://drive.google.com/uc?export=download&confirm=1&id=1d3wPbtjFcgCkWAMVFQalOuQHdiNmfc5i
4xFaceUpLDAT (for lower quality input, can additionally denoise)
https://drive.google.com/uc?export=download&confirm=1&id=1Bc6kw0c-axNL5vCdtLMgz8fRnvJWcgVn
4xFaceUpSharpDAT (for good quality input, produces sharper output)
https://drive.google.com/uc?export=download&confirm=1&id=1idWh9oHSfzTzR_-2UVjd22Rb7ihJApJf
4xFaceUpSharpLDAT (for lower quality input, produces sharper output)
https://drive.google.com/uc?export=download&confirm=1&id=1Bc6kw0c-axNL5vCdtLMgz8fRnvJWcgVn
8x_NMKD-Faces
Purpose: upscale images and enhance face details
Date:2023-07-10
Download:
8x_TGHQFace
Purpose: upscales blurry 128px faces, usefull for enhancing that someone in a small picture.
Scale: 8X
Date: 2019-11-27
Dataset: flicker cropped face
Architecture: ESRGAN
Download:
https://drive.google.com/uc?export=download&confirm=1&id=1OyOJIW224hBhb-aTCbuUQb0qzKmE4oH6
4x_Face-Ality V1
Purpose: upscale small faces to big faces
Scale: 4X
Date: 2019-11-09
Dataset: Lots of faces, mostly real life but some anime.
Architecture: ESRGAN
4x_BigFace
Purpose: upscale images of artwork and people
Scale: 4X
Date: 2020-07-24
Architecture: ESRGAN
Available versions: BigFace, BigFace V3, BigFace V3 Blend, and BigFace V3 Clea
Download: https://1drv.ms/u/s!Aip-EMByJHY27CHg3-1Uue5KgbtK?e=6k0COU
4x_FArtFace
Scale:4X
Date: 2019-11-28
Architecture: ESRGAN
Download: https://1drv.ms/u/s!Aip-EMByJHY211Vcr3fF39ZOaPk-?e=BQajTd
4x_Face Focus
Purpose: face de-blur - slightly out of focus / blurred images of faces. It is aimed at faces / hair
Scale: 4X
Date: 2019-05-19
Architecture: ESRGAN
Download:
https://drive.google.com/uc?export=download&confirm=1&id=19ICMKNuS4PbhmtA7Be9lUE6v8NHDDEzP
4 Upscalers for Text
Text2HD v.1
Purpose: upscale text in very low quality to normal quality
Scale: 2X
Date: 2024-08-27
Dataset: scanned books and text
Architecture: RealPLKSR
8x NMKD Typescale
Purpose: upscale low-resolution text/typography and symbols
Scale: 8X
Date: 2020-11-04
Dataset: custom - typography images from flickr
Architecture: ESRGAN
Download: https://icedrive.net/1/43GNBihZyi
4x_Sourcetex
Purpose: upscale of textures, especially for grungy/dirty/rough ones
Date: 2020-10-25
Architecture: ESRGAN
Available variations: 6 (Clean/Anti-Artifact with 3 smoothness levels each)
Download:
2x_BSTexty
Purpose: upscale text with less distortion that other models. It seems to do a good job generally, but don't expect it to be a state of the art model that can upscale magazines and stuff. It makes things more readable but since it was train on B/W pictures it desaturates them
Scale: 2X
Date: 2020-03-04
Dataset: Randomly generated text, with randomly chosen fonts with varying sizes. Black text on white background
Architecture: ESRGAN
Download: https://drive.google.com/uc?export=download&confirm=1&id=15ovbadCoYs7q8nSd5Mq02PqBOpwiBkoS
Use the Models
Step 1. Download and install AUTOMATIC1111’s Stable Diffusion WebUI by following the instructions on https://github.com/AUTOMATIC1111/stable-diffusion-webui
Step 2. Place your model file (ending in *.pth) into the folder of the corresponding official model it was trained on.
For example, if you downloaded 4x_Nomos8kDAT.pth, put it in the /models/DAT/ folder.
Step 3. Launch WebUI. Go to the Extra tab and drag your image into the uploading area.
Step 4. You can then choose the upscaling size in 2 ways:
Scale by: Choose a multiplier (e.g., 2x to double the size). The default is 4x.
Scale to: Enter specific width and height values if you prefer exact dimensions.
For this guide, we will use the default 4x scaling option.
Step 5. Choose the model you just added.
Step 6. Hit Generate and we can start the upscaling process.