I take it the missus knows your turning her into a witch 🙂
I take it the missus knows your turning her into a witch 
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Sshe does lol
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dbrave guy
Swe'll probably do an actual photoshoot once the weather and leaves change
SDlol
Sshe loves halloween
Dmakes sense now
SAI makes for good inspiration at the least
Dso you gonna be a wizard>?
Sno she will probably make some something i'll hate like prince charming or shrek
Dheheh
Dgotta post them mind
Si've spent most of the weekend training to try and get good custom results that are realistic but not adding all the gross witch feature 
Sshe wants to be a pretty witch not a gross old wrinkled one
S
Sthis one is getting there
Sstill some artifacting though
Dyeh stay clear of the warts
Dif you know whats good for ya
Strying to decide if i should be using file words or not for the witch concept
Si think those ones had better results
Sbut they are less creative
Dno green skin?
S god noDlol
Sis inpainting the only way to get a face to show up well on a full body shot?
Dhmm yeh tricky when faces are further away
Dtends to need a little help
Swell that's why i was thinking if i could to a high res fix at 4x maybe it would work
Si'm also curious to why i can do text to image 4x hires fix but if i put a 768x768 image into img2img and upscale that much i run out of vram -_-
Seven though i can do it with 4 at a time in txt2img
Dwhat resolution is the text to image
S768
Slike i could do all 4 images 768x768 and hirex fix at 4x but if say i just want to make more generations faster and i do no hires fix then find the ones i want to upscale with img2img i can't go that high
Sso i have to upscale by maybe 2 or 3 then use the extras tab
Sbut i can try that video he linked and see how it does
Dgive it a go
Sprobably try rendering at 1080x880 so it's more of a printable size
Si was trying to get maybe some spell casting photos but could never get the hands right
DI don't think I agree that LR 8e-6 is better than 1e-5 on OT, on my test dataset. some more skin details on 8e-6, but likeness is reduced
Flower LR means more training necessary but not 1 for all 
Fdepends on dataset
Fi try to balance
Dthe best one I've found is a cosine wave between 0.5e-5 and 1.5e-5. it's got the best of both, skin details and same likeness as 1e-5. that's not so easy to configure though
Dand I haven't tried any other values. it was just an idea but it's surprisingly great
Fsurely configs can be specialized
Fi am trying to make config that wont very undertrain or overtrain on every dataset :d
DI don't know yet, but a wave around a midpoint might even be more general for every dataset than a fixed value. I don't like the cosine annealing with restarts and such, they waste to much steps at the bottom. but this one seems to work because it stays mostly around the LR you want
Fthe config gives very good likeliness and generalization between 100 epoch and 300 epoch that means flexiable
RHello I have a question what is the difference between dreambooth and finetuning in koya. If possible I want to know the details.