Ok I'm patreon supporter now:)
Ok I'm patreon supporter now:)
Fty please open google remote desktop
Fi will quickly install
Falso start a fresh runpod instance
Fi train a lot of models, using different dreambooth options and when i want to compare models to get the best one..... it's an impossible task 
Fdepending on seed, cfg, etc... sometimes a model is better at 3.000 steps, other times at 20.000 steps, other times using filewords, other times not using it...
Fdamn it
Fwhat a waste of time
Fi have like 40 models of the same thing and can't decide which ones to delete
Fi also tried using deepface to get the scores but it also depends on many things!
Fwhat is your ultimate aim?
Fmy results are not even consistent, even if i'm using the deterministic option
Ftrying to get a model or two, one for best txt2img, flexible and good quality, and if needed, another for inpainting with insane quality
Fmaybe i shoauldn't be using euler_a to compare? when comparing checkpoints at different steps... they produce very different stuff and it's hard to compare
Ffor example model a 13920 steps produces a totally different image at 17400 steps
Fso it's impossible to compare at what epoch is better, because results are so different
FYou are using same training dataset?
UYeah there's no way you're guessing what cpu I have....
UAnd 64gb more ram in the mail 
U(totally replied to the wrong sentence, oops)
Fyes
Fnice
Fwith deterministic it is suppsed to be same but as you say it is not
Fthere are random parts of training and i think they cant make it deterministic
Fespecially with optimizer if you use
Fit's very hard to compare stuff
Fi tried 10 class images vs 30 vs 50
Fand i can't see a difference
UClass images? If using class images, aren't you supposed to have significantly more? Last time I used class images in training it was smthg like 100 per training image, and then 100 or 150 steps per epoch. Oof I'm a little rusty off the top of my head.
Tinterestingly it actually does in chatgpt4 since you can give it internet access. i didn't choose to use it in this case though
UChatGPT 3.5 and earlier only has knowledge up to September 2021.
Tah yep true
Thopefully they'll make gpt4 free for everyone eventually. i think it's worth the $20 a month though tbh
Tit also has plugins now for various things like the stockmarket and tutoring
MIs there a way to see the preview image?
Fi think there is but i also dont know yet :d
Tput the image you want as the preivew into the same folder as the model. the picture must share the same name as the model and be a png file
Fah not auto generating
Fyou put them manually
F
Fi am going to sleep see you tomorrow hopefully
DHi!! I was setting up my stable diffusion today but for some reason I'm not getting my local url, was wondering if anyone would be able to help?
Pwdym? it should just be localhost:7860
DCan I DM you what I mean?
DNevermind, I think i'm ok, I just re-ran it and it worked
KHave you done it yet? I'd love to see what you have found out.
KAlso, I found your videos on many GitHub pages, such as this one: https://github.com/bmaltais/kohya_ss or Runpod page. I don't think other YouTubers are featured as frequently as you, which demonstrates how high your credibility is.
GitHub
Contribute to bmaltais/kohya_ss development by creating an account on GitHub.
AThat doesn't really look like it respected the prompts too well, unless that was some checkpoint that has such specific style and no loras.
AI tried regional prompter and while it worked somewhat with the horizontal/vertical divides it quickly failed when I tried it with masking, especially since I was trying to use loras for specific parts of the regions
Awhen I tried latent instead of attention mode (which is supposed to separate loras into regions) it rendered something that looked nothing like what I was getting withotu regional prompt or with regional prompt in attention mode
AAnd yeah, I'd rather avoid using inpaint since the goal is to generate lots of decently looking images straight out of the txt2img with maybe optional pass of batch img2img to refine it or alter the style somewhat.