>>3009023Yeah that's the bare minimum nodes, plus a bunch of loras.
(left - base res)
Now you might want to upscale and run a second pass, that was the top right corner labeled Hiresfix. AI can't make good detail without more resolution, but at the same time it cannot make good anatomy if going above 1 megapixel. So we use a 2-step process, generating the composition and anatomy first, then upscaling and doing a second AI pass at low denoise. This improves details, and the low denoise prevents it from changing anything major such as breaking anatomy. You can use the Latent upscale node from your workflow (but Latent requires high denoise to cover up upscaling artifacts), or better grab an upscale model from
https://openmodeldb.infoNodes for that would look like this.
https://litter.catbox.moe/fukjy22bwnfeucnf.png(middle - after hiresfix)
And then optionally further refine the face with the Face Adetailer section you had at the bottom. It does a crop+upscale of the area (auto-detects where the face is) also just to give the AI more resolution to work with (and not have to upscale the whole image). Alternatively you can load the image back in and draw the detailer mask manually, like I do at the bottom. That way you can improve whatever area you want and not just what you have a detection model for. And again you have a denoise level here that determines how much the area is allowed to change.
(right - fixed eyes)
