Favorites #

On this page I present my favorite upscaling models for a given example together with a simple recommendation. If you want all the upscaling outputs of these examples, head on over to the multimodels page.

Example Controls: Left mouse button to drag the image or to move the slider, mouse wheel to zoom in, right mouse button to toggle left model on/off, releasing middle mouse button will activate a short flicker test for the left side of the slider. Do not work on mobile.

Buddy #

For photos with faces my simplest recommendation is SwinIR-L together with CodeFormer in chaiNNer as shown here.
GFPGAN and CodeFormer results can also be blended, I included the 4xLDSR_blended output as an example (50% visibility each with GFPGANv1.4 and CodeFormer with 0.7 fidelity) as shown here

Input4xLDSR_50steps+CodeFormer4xLDSR_blendedBSRGAN+CodeFormerRealESRGAN_x4plus+CodeFormerReal_HAT+CodeFormerRemacri+CodeFormerRudalle+CodeFormerSwinIR-L+CodeFormerSwinIR-L+GFPGANv1.4SwinIR-LUltraMix_Balanced+CodeFormerrealesr-general-wdn-x4v3+CodeFormerInput4xLDSR_50steps+CodeFormer4xLDSR_blendedBSRGAN+CodeFormerRealESRGAN_x4plus+CodeFormerReal_HAT+CodeFormerRemacri+CodeFormerRudalle+CodeFormerSwinIR-L+CodeFormerSwinIR-L+GFPGANv1.4SwinIR-LUltraMix_Balanced+CodeFormerrealesr-general-wdn-x4v3+CodeFormer

Drop 1 or 2 images here !

Grosser Mythen #

For photos of landscapes my simplest recommendation is Real_HAT_GAN_SRx4 with chaiNNer.

LDSR also gives good results here but is not supported by chaiNNer. If you are using LDSR (with Automatic1111 or Replicate) always check your output for completeness. Oftentimes the input is not in the correct dimensions and therefore unintentional cropping will occur, then you need to manually pad the input and crop the output.

Input003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN4x-UltraMix_Balanced4x-UltraSharp4xJaypeg904xLDSR_200steps4xRealSR_DF2K_JPEG4x_NMKD-Siax_200k4x_RealisticRescaler_100000_GBSRGANRealESRGAN_x4_rudalleRealESRGAN_x4plusReal_HAT_GAN_SRx4realesr-general-wdn-x4v3Input003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN4x-UltraMix_Balanced4x-UltraSharp4xJaypeg904xLDSR_200steps4xRealSR_DF2K_JPEG4x_NMKD-Siax_200k4x_RealisticRescaler_100000_GBSRGANRealESRGAN_x4_rudalleRealESRGAN_x4plusReal_HAT_GAN_SRx4realesr-general-wdn-x4v3

Drop 1 or 2 images here !

KonoSuba #

For anime images my simplest recommendation is RealESRGAN_x4plus_anime_6B with chaiNNer.

Input003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN4x-AnimeSharp4x-UltraMix_Balanced4x-UltraSharp4xLDSR_50steps4xRealSR_DF2K_JPEG4x_NMKD-UltraYandere_300k4x_foolhardy_RemacriBSRGANRealESRGAN_x4plus_anime_6BInput003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN4x-AnimeSharp4x-UltraMix_Balanced4x-UltraSharp4xLDSR_50steps4xRealSR_DF2K_JPEG4x_NMKD-UltraYandere_300k4x_foolhardy_RemacriBSRGANRealESRGAN_x4plus_anime_6B

Drop 1 or 2 images here !

Fate #

For anime images with bokeh effect my simplest recommendation is AnimeSharp with chaiNNer.

If only a 2x upscale is needed, you can also try out 2x_Bubble_AnimeScale_SwinIR_Small_v1 with chaiNNer.

RealESRGAN_x4plus_anime_6B this time does not belong my favorites, I left it in here for you to notice what it does to the blurry background.

Input4x-AnimeSharp4x-UltraMix_Restore4xRealSR_DF2K_JPEG4x_BooruGan_650k4x_NMKD-Siax_200k4x_NMKD-UltraYandere_300k4x_NMKD-Yandere4_120000_G4x_NMKD-YandereNeo-F64B2_200k4x_foolhardy_RemacriRealESRGAN_x4plus_anime_6Brealesr-general-wdn-x4v3Input4x-AnimeSharp4x-UltraMix_Restore4xRealSR_DF2K_JPEG4x_BooruGan_650k4x_NMKD-Siax_200k4x_NMKD-UltraYandere_300k4x_NMKD-Yandere4_120000_G4x_NMKD-YandereNeo-F64B2_200k4x_foolhardy_RemacriRealESRGAN_x4plus_anime_6Brealesr-general-wdn-x4v3

Drop 1 or 2 images here !

Life #

For AI generated images, my simplest recommendation is Remacri with chaiNNer.

Input003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN4x-UltraMix_Restore4x-UltraSharp4x-UniScaleV2_Moderate4xLDSR_200steps4xZZines4x_NMKD-Siax_200k4x_RealisticRescaler_100000_G4x_foolhardy_RemacriBSRGANInput003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN4x-UltraMix_Restore4x-UltraSharp4x-UniScaleV2_Moderate4xLDSR_200steps4xZZines4x_NMKD-Siax_200k4x_RealisticRescaler_100000_G4x_foolhardy_RemacriBSRGAN

Drop 1 or 2 images here !

ColorJacket #

Another 'more artsy' AI generated image, my above recommendation still stands and stems through looking through my ai generated examples from the multiple models page (not just this one or the previous example). I simply wanted to show that depending on the generated image, you might also want to try one of these models instead. For example, I had used the RealESRGAN_x4plus_anime_6B model in the past to upscale generated logos.

Input003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN4x-UltraMix_Restore4x-UltraSharp4x-UniScaleV2_Moderate4x-VolArt4xFSDedither_Manga4xFSMangaV24xFatalPixels4xFatalityV14xFatalityV24xLDSR_50steps4xZZines4x_BigFace_v3_Clear4x_Fatal_twerpilation4x_HDCube2_250k4x_Unholy_FArt4x_foolhardy_RemacriRealESRGAN_x4plus_anime_6BInput003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR-L_x4_GAN4x-UltraMix_Restore4x-UltraSharp4x-UniScaleV2_Moderate4x-VolArt4xFSDedither_Manga4xFSMangaV24xFatalPixels4xFatalityV14xFatalityV24xLDSR_50steps4xZZines4x_BigFace_v3_Clear4x_Fatal_twerpilation4x_HDCube2_250k4x_Unholy_FArt4x_foolhardy_RemacriRealESRGAN_x4plus_anime_6B

Drop 1 or 2 images here !

Denoising #

My personal favorite would be SCUNet which can be used for example with replicate. As an alternative with chaiNNer you could use the SwinIR denoise models. The examples are in the denoise page.

Deblurring #

My personal favorite would be MAXIM which can be used for example with replicate. As an alternative with chaiNNer you could try out the 1x_ReFocus_V3_140000_G model. The examples are in the deblurring page.

JPEG Artifact Corretction #

This one is harder, I think my personal favorite currently would be FBCNN which can be used for example with this huggingface space. As an alternative with chaiNNer you could for example use one of the 1x_JPEG models, one of the SwinIR colorCAR models, or even the Swin2SR_CompressedSR upscale model and then downscale the output back to its original size. The examples are in the artifacts page.