>>2772627Welcome to /sdg/
Generally stick with the vae that was recommended with the model. Also if you take a look at the file hash you'll find that most of the vaes are actually the same one.
I recommend perusing the UI wiki because it has a lot of good stuff about how prompts work.
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features>I read something about prompts near the beginning being more strongly valuedIt's true, order matters. However, you can sort of override this using emphasis (see the wiki). I typically start the prompt with whatever primary character or characteristic I'm going for.
>Quality tags like masterpiece, or those relating to the character?The masterpiece trick stems from the NovelAI leak about 6 months ago. It was discovered that part of their secret sauce included putting "masterpiece" at the beginning of every prompt. I never use it, but a lot of people like the style it outputs.
As another anon mentioned, use lower base resolutions + upscaling or highres fix. The reason for this is because the models are trained on 512x512, 512x768, 768x512, and in some cases 768x768 inputs. You'll have best results if you stay in those bounds (you don't have to keep those exact resolutions, just somewhere in the 768x768 box), otherwise you'll experience tiling and other weird results.
>but is there a specific trick/tag for the faces?I found that just mentioning eyes causes it to improve the eyes and face quite a bit.
