Not sure whether I should post a reddit link but this guide is excellent.
Stable Diffusion
Discuss matters related to our favourite AI Art generation technology
Also see
Other communities
Blasphemy!
Thanks, looks useful.
This guide is ok but SD Upscaler is outdated.
After refining to 2048px with img2img I usually send it to extras and upscale to 4096px with 4x-UltraSharp but for the most natural looking result I will just scale the 2048px image up in photoshop and then overlay the UltraSharp output somewhere between 40-70% opacity. But then it also depends on the style of the image, for some images that I really want to look their best I'll do both UltraSharp and R-ESRGAN-x4+_anime_6B and then mask out the areas I find to look best between the two in order to get the fewest artifacts. Maximum sharpness isn't always the best, some mild blur and even film grain/noise can do a lot to make an image look really natural and beautiful.
Better than ultimate SD upscaler: https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111
Combined with controllnet tile upscale it provides amazing results.
So I did install this but couldn't find "Tiled VAE" or "Tiled Diffusion" anywhere in the UI.
Is controlnet something I have install separately?
Yes you have to install ControlNet separately.
But you should already see the Tiled VAE and Tiled Diffiusion plugin.
Watch this video to see how it should look. Also a guide for this method: https://youtu.be/KkxNEiiBOyQ
Oh I do have them. Was looking on Extras with the upscaling settings.
Watch the video for settings. It is quite complex. If you want to add detail or change the image more try ControlNet+Ultimate SD Upscaler script. It is also easier with the settings: https://stable-diffusion-art.com/controlnet-upscale/#Method_3_ControlNet_tile_upscale
Okay, thanks for the help