Dude all of my images are coming out looking like some mutated 12 armed bodies that picaso painted... Nothing looks even REMOTELY realistic. No matter what my prompts. Or my negative prompts.
I appreciate this. I keep seeing one that uses anti-immigrant rhetoric. I work in immigration advocacy so that especially pisses me off. A lot. I'd love to do a deep dive on these freaks, even talk to them, but that's just my morbid curiosity. Still, angry.
17:40 always get a gaming laptop instead of a office one if you can afford the extra electricity 500$ will get you WAY more preformance with a gaming one even if you never game.
I think it sounds good, but maybe work on muting the strings. Also the canera mic isnt helping, I think that's what people are getting it sounding awful from
@@heggy_69 it's your 2nd comment. I had the Behringer set for bass and forgot. I did the recordings through a mixer to line in on a discreet sound card
LOL I have a box of the common screws for computers... This video was 4 years ago... But I do know that SSD's and hard drives take the same size screw, and it's standard.
Regarding the seeds in comfy vs. elsewhere, I think they do noise generation a bit differently so the base latents start out differently, but I know for sure that comfy does not do token weighting normalization that a1111, and forge do. So Identical prompts pass different vectors into the model.
Not too long, but maybe something to try in the future if you get a low view count is to break it up into 2 or 3 parts. You have about 3 major topics you segmented this into that could each be their own vids. But I didn't mind the longer runtime, although I did put it off for a day that I had more time to view it all. Just a suggestion.
Dude! You are a god send! I tried so many otehr posts to get Forge to recognize my Automatic 1111 directory and your method was the only one that worked.
Amazing video! Been so drown into that new model that honestly gave up figuring out how body works for 1.5 and SDX.L. I always I wanted to train a korean singer that doesn’t fit that korean standard of super skinny with small bungaroos 😂 I made like 20 loras, having a good face but the body sometimes was decent, some others it was not-at-all. Need to try the crop trick + making and independent folder for the body 🤩 and see if that helps. Happy to always find useful information in your videos!
@@milycortes3100 try adafactor for both the optimizer and the scheduler, have it output every 100 steps instead of x epochs That's a tip from a future video lol
Thank you for the research and thank you for sharing the cropping trick idea for training! How do you caption upper and bottom half though? Do you just use the same captions for both images or do you tag them differently?
Okay So here's the thing , I've been watching your videos for a while and I personally think you are one of the best out there with so much attention to detail. Anyways ,Before creating a lora i found one on civitai and thought id test with it first but everytime I try to generate NSFW content and add the lora but it never appears,The face never matchs the face of the lora ,like its not even close. Like it legit never works, I've tested it with a few different checkpoints like EpicRealism, PonyRealism and Acornisspinning...Is it even possible to do it? I've even tried using Chatgpt for prompts but Its impossible..
@@damianlazo5760 I cropped so people can see what I am clicking on, I have the problem where I can't see what the person is clicking on because they're monitor resolution is so high