Try this one https://huggingface.co/nyanko7/flux-dev-de-distill its behavior is completely different
Try this one https://huggingface.co/nyanko7/flux-dev-de-distill its behavior is completely different
Dthey did something to it, may be it surprises you
Ci know what they are doing, and it's not what they think they are doing.
Cplus - it's 100% impossible to put information back that's been removed if you don't know what it is - and they don't have the original weights.
D@Dr. Furkan Gözükara Kohya updated nyanko7/flux-dev-de-distill is training now.
DI agree you cant get that was deleted back, but it doesn't mean that you can't put new information on its place
Cbut that doesn't undo what was done, does it?
DI agree
Cif you have a bag of chocolate chips, i take them out, you have an empty bag. if i then put potato chips in it, do you have a bag of choclate chips again?
Cand even if i put more chocolate chips in it, do you ahve the same chips you had?
DWe want a functional model I dont care if it is identical to the original if it works
Dflux pro is private and we will never have it
Cyou'd be better off just using SDXL - what was done to flux was not only the distill process, but other things as well. those were done to try to fix an issue that the neural network has. that's resulted in flux being rigid, untrainable, and giving very strange results outside of a narrow window.
Cand the licenses do not allow you to modify flux the way that some people are trying to do
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
Cflux by itself, if prompted correctly, doesn't need loras or fine tuned for any reason other than the very specific issue of you wanting to make images of stuff it can't possibly know about, anyway
DIDK I'm training a lora with many subjects on flux-dev-de-distill right now, kohya was updated some minutes ago and now supports the modified model
Dlets give it a chance
DI use SDXL too, but the quality is much worse
Cthere's no reason to train loras, or make fine tunes of flux, for any reason other than needing to be able to generate images of something it has no data at all about. the community, for some odd reason, thinks they can't generate without all those add ons. all it takes is learning how the model thinks, and then prompting it correctly
Fflux is perfectly trainable for my purposes 
Fthat is all it matters
Cother than to plaster your face all over social media, what puropose?
Fwhatever the purpose you have, you can trian your style generate images with it
Ftrain object
Fitem
DThe purpose of training or finetuning is to add new concepts people or things that the model don't have, what other reason you can have?
F100% that is the aim
Cbut you never do anything but pose images. we tried yesterday to get you to do fun, unusual images.
Fnope i can do all poses i need
Fi can even surf
Flol
Ci want to see an apple with your face
Fi dont wanna see it
Fsimple as that
Csee? boring pose imges
Fnope
Fi got all the fun and images i need 
Fi dont see any fun in such image
Fif i can ride a dinasaur an eagle that is more fun
Fbut that is not the aim
Fas i said
Ftrain object style item
Fthese are more important
For get professional photo for linkedin and such
Fthe only lacking thing is stylization atm
Fi hope we will also solve it eventually
Fit is poor
Cnot if you keep trying to train an untrainable model, you won't
Fwell people trying distillation i cant say for sure will work
Fbut we will see
