New Home Base for Image generation (Hint: Big performance increase!!!) (Patreon)

So recently I discovered a alternative version of automatic 1111 web UI, called A1111 Forge!
It's created by lllyasviel who has created some awesome ControlNet models for Stable diffusion and other awesome stuff in the space!

It's a rebuilt version of A1111 but without all the unnecessary stuff thrown in and changes that were opinionated.

And I started using this new one Just recently after I did some of the works already.

And the performance is much much much much better!
I'm not even using any command line arguments to help with performance as well and I always used them! I always needed Medvram and xformers every time!

One of the big problems that I've had using the other version of automatic 1111 was using high res fix, This process has always been a bitch for me to deal with because it usually takes a while to run for each image and it adds up. I especially needed it for original anime look of characters aka anime screencap which is terrible on hands, they always looked off and had artifacts a lot of the time so I tried to avoid this alot!

When I tried this new UI it was incredibly fast and I was able to make even larger images.
Doing high res fix for larger sizes means better quality and generally clearer and better hands! Also breasts essentially not need to be touched up(I still use it tho). 

Usual Image generation size 640x790 then usually did 1.3x upscale to 832x1027(2 minutes to complete on original A1111, and now 1.15 minutes), now I can make 960x1185 (now do 1.26 minutes to complete). So expect generally better results and cleaner anuses! xD (Cloud member see my Megara and Momo reverse cowgirl sets I posted today on the cloud), Momo was on anime screencap model which usually just looks bad, and Megara is on my main model!

When an image is small and depending on how the character is posed hands usually end up looking pretty blurry or just not that good. If they're pretty bad then I can't even use a detailer for the hands, because the hands are so messed up It can't identify where the fingers are.

I would say the performance increase on this thing is pretty much about almost two times the speed, maybe even more because I can do even larger sizes than I usually would never touch!

Also when images are done in between normal generation and hires fix and after highres fix it's much faster!

When I tested out stable diffusion XL and with high res fix which I never usually use because it's so slow, I'm able to generate images way faster than I've ever could have compared to the other UI. Note SDXL is a 6.5GB model alone and I have 8GB Vram!

This opens doors to making NSFW content on SDXL which is a big improvement on accuracy and prompt articulation!

The only issue which is actually getting fixed and he's currently working on it is dynamic Lora which I'm sure some of you guys have heard of through me.
It's literally the backbone of my works and I always usually use it so that characters are posed the right way but also show up properly still.

Comfy UI is also great on performance but dynamic lora and BREAK keywords are not there due to different syntax and handling of models.
SD Next, is more so SD Last Imo, I have nothing but bad experiences with it, sure performance might be ok, but the whole thing is just very confusing and tons of unnecessary things you got to do, to get something to work. Even removed some crucial things I relied on as well just because of some dumb reason which I never had issues with.
Invoke AI.......yeach F it, I CAN NEVER GET IT TO WORK, every time I try loading it up it constantly scans all models and asks you to set some setting on each one and have to do it again each time you load up.

lllyasviel has been around for quite a while and has made amazing things for the Stable Diffusion community, so him making a A1111 Webui of his own was a big deal for me and the amazing performance of this is already showing how good he is at what he does!

This is why I'm such a MAJOR fan of Stable Diffusion over any other image generators, due to the insanely big open-sourced community, there are ALWAYS new tech and techniques, new kinds of models and all sorts of things being released thanks to the awesome research computer scientists do,