There are two basic negative prompts I default to in producing anything. Add them first and dedupe against any of your potential positives:

low contrast, ugly, deformed face, fat, obese, disfigured, deformed, deformations, blurry, out of frame, long neck, long arms, big ears, bad hands, zoomed in, deformed arms, extra limbs, extra nipples, extra legs, extra arms, extra fingers, amputee, pants, bad anatomy, bad proportions, mutation, furry, mutated, extra_limb, penis, text, watermark, male, missing limbs, distortion, black and white, low resolution, missing hands, distorted hands, low detail, closeup, censored, deformed hands, crossed eyes, out of focus, bad framing, tiling, doll, grainy, cat, fashion, shoes, grayscale, monotone, old, blind, amputation, smooth, watermarked

(((duplicate))), (bad legs), one leg, (bad face), (bad eyes), ((bad hands, bad anatomy, bad feet, error, missing fingers, fewer digits, cropped:1.0)), worst quality, low quality, normal quality, jpeg artifacts, signature, ((watermark)), (username), blurry, out of focus, censorship, ugly, old, wide face, ((fused fingers)), (((too many fingers))), amateur drawing, odd, fat, lowres, text, error, out of frame, (cloned face:1.3), (mutilated:1.3), (deformed:1.3), (gross proportions:1.3), (disfigured:1.3), (mutated hands:1.3), (poorly drawn hands:1.3), (bad hands:1.3), extra digit, (extra fingers:1.3), (poorly drawn feet:1.3), long neck, extra limbs, broken limb, asymmetrical eyes, cell shading, watercolor, monochrome, artist name, anime, cartoon, bubbles, long torso

Also add whatever general prompts you want and add in the improved prompt matrix to make life easy. then on a specific key phrase add in 'PHRASE:<0.1|0.5|0.8|0.9|1|1.05|1.1|1.15|1.2|1.25|1.4|1.45|1.5>' in order to experiment with how far you can push the model.

Once these are solidified, key them into your base prompt dictionary for future generation. Once a solid base is built, I like to feed them with reference videos into img2img to make .webms for /gif/. It takes time and a lot of trial and error, but the base are use the negatives and for positives, go ham.

dude has two children in his prof pic and wants to know how to make sex pics... suss

click at the little i in the botomm left of image uploads. you get the prompts there

vae-ft-mse-840000-ema-pruned

Dude likes old, fat chicks. Go figure.

sheeeit

Retard.

@Reidrigo Purulent asshole.

@jdsguam Nah, just really disgusted with the child porn.

Stable Diffusion Webui loads safetensors by default if placed in the models folder.

Make sure you have the latest version of A1111, place into the model/stable-diffusion folder and add set SAFETENSORS_FAST_GPU=1 to your webui-user.bat

@MarcusVerus What does "SAFETENSORS_FAST_GPU=1" Do exactly? What if my video card is older, like a 4GB card? Will that still help or make it worse?

it depends on the notebook.

I totally agree. Upscaling is already here, but I need realistic hands. You can head over to https://arc.tencent.com/en/ai-demos/faceRestoration to fix your eyes and lips. You also get an automatic boost in resolution when you download your sample.

actually, Hassan has some great hypernetworks that really boost the quality of fingers, limbs, poses, etc. I highly recommend his work in using PFG.

@beefwittedbuttocks Thank you. I'll check them out.

I'm a little late to the party on this comment thread, but I use ADetailer to do fix ups for this kind of thing. https://github.com/Bing-su/adetailer This is a great replacement for face restore and helps with bad hands too.