>>2771111I can give you the settings I use, along with the VAE and favorite models.
I always use the DPM++ 2M Karras Sampler, works best from my experience at 30 steps as the sweet spot. HiRes Fix is also a must if you have at least 8Gb of VRAM, with the R-ESRGAN 4x+ Anime6B upscaler with at least 5-10 steps. The upscaler really cleans up some of the noise and blurriness, I know some folks will tell you to just send images to extras to upscale but I believe that will only go through 1 step, so you end up with images that are more blurry when you zoom in. I leave Clip Skip in the settings to 2.
As for models, you can search these on CivitAI, my favorites are:
CetusMix - Really good brush strokes with pastel like art.
ExpMixLine- Really good solid lines if you want more manga-esque images
AnythingV5.0- The most updated Anything model supposedly made by the original Anything author, V4 and V4.5 are joke mixes made by another person.
MixProV4- A good general purpose model with clean lines and eyes.
Also as an addition, I would download text embeddings, most popular are easynegative and bad_prompt_version2, which basically adds a bunch of negative prompts into one package.
Most importantly is the VAE you use, since SD sampling washes out colors and you get these desaturated, blurry messes. People will tell you to use the AnythingV3.0 VAE but the one I use exclusively outside of baked in ones models use is
https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckptfor really crisp images.
