how many input images did you use?
how many input images did you use?
F
Fi did 2 testing
F1st case me : 15 images ohwx man, me 15 imges man
F2nd case me : 15 images ohwx man, 5200 reg images man
Flet me show
Fboth 15 images
Fohwx man vs man
F
DF
Dif you've used bs > 1, the code doesn't work with that
Fdmn
Fyou are right
Fi made it 1 sample
Fnot 15 :d
Fin reg images
Dor could it be that there was mistake in REPEATS and you've used 15 trainimg and only 1 regimg?
Fye will train again
Fso in my reg images training
Fit used 1 reg images
Finstead of 15
Fat every epoch
Dthere go another 8 GPUs 
Fgood catcgh
F
Feducated eyes difference you caught that 16 step very well
Fstarted another training
Dwelll...guess who made that mistake before
Dhave you tested this one? there everything looks fine. about 15 reg steps out of 30
Fyes tested
Fyou can see above
Fit is me twice
Fso me twice is
F
Fmy images twice
F1 ohwx man
F1 man
Dthat should work though 
Fwell we got loss rate too
Fbut didnt work
D3rd image shouldn't be you then if I understand correctly
Fyep
Fthis is just a man prompted
Dif you wanna investigate, I'd just use "man" and "ohwx man" as X on a grid, and checkpoints on Y. 4500 is a lot and maybe you are using stronger parameters than I, and it worked on the lower checkpoints? just a guess
Fcurrently i am testing on 2250 steps, 150 epoch
AHmmm, interesting. A question about the fine-tuning of checkpoints: can I “retrain” a checkpoint without losing the quality of existing people? Is everything re-evaluated when this is done? Thanks in advance!
AWith checkpoint finetuning, is it possible to have more than one "named" trained person alongside in a render image? if I do this within one lora, all people look the same^^
CI dont believe that is possible unless you are training the checkpoint on a new concept/class, if you want to train a new person/man/woman etc it will overpower existing data. The reg image experiments they are trying above is to attempt to stop the same person issue, but so far no one has been able to solve it and it would ultimately just be training new generic people not retaining existing people
ASo in this case I have to add all persons at once?
A(or objects, etc?)