for Flux Dev model, the best setting we already have, just use @Dr. Furkan Gözükara file
for Flux Dev model, the best setting we already have, just use @Dr. Furkan Gözükara file
V@hamuz I am using this one and enjoy the results. https://medium.com/@zhiwangshi28/why-flux-lora-so-hard-to-train-and-how-to-overcome-it-a0c70bc59eaf
OIt seems like, for now, if you are only training one subject, the dev model is probably your go to.
If you want to try something fancy, creating a full finetune checkpoint then extracting the LORA has given me my best LORA's
If you want to try something fancy, creating a full finetune checkpoint then extracting the LORA has given me my best LORA's
HAlright, thanks man! I’ll train a few different LoRAs with different settings, and I’ll share my experiences here with you. Maybe after doing enough research and gaining some knowledge, I’ll look into the distilled and de-distilled versions
Wthis one I dont see anything different compare with Flux dev Lora
Wjust create different image only
WPeople also say the same
Hofcourse I will 
V@wardensc2 You tried both? Which output was more aligned to what you used for data?
Wyes I try both
Wcompare 2 loras: Flux and that model
Wjust different image only
Wnothing improve
Wpeople suspect it just some kind of merge checkpoint and pretend something new
HAlso, has anyone here rented and used an H100? Is there really a significant speed increase in training? Compared to 4x A100
VThere is also the method that ai toolkit uses to train. Some sort of addon model that is based on schnell. A few other distilled as well.
WTo be more specific I train 2 loras with about 30 images and about 5800 step, result is no improvement
VSame config?
Dflux-dev-de-distill is the best for training I'm very impressed, then you can use the lora on regular Flux-dev any quantization as normal
Wyes
Wjust change the checkpoint only
VMaybe it needs different config since the data is distilled.
Dthe best option will be to full finetune flux-dev-de-distill and then extract the lora
VI want to do that, but I doubt my 4090 can do it in a reasonable time.
Wdo you read the author said
Whe said he finetune with 3 million high quality data
WI bet he need to have a company like facebook or amazon
Wto do that
W= )))
WI dont buy it at all
Dthe config is the same for training, is different for inferring it uses real cfg rather than Distilled CFG Scale so is slower to generate pictures
Wbase on how long to train just 30 images for a checkpoint
Dfor that reason I use the trained lora on regular flux-dev
VYeah I would like to extract the lora.
D
Wtraining checkpoint of flux is require much better hardware than lora, so training a checkpoint then extract a lora is not an efficient way to get a lora. I rather train a lora directly
V@wardensc2 I tried the 4090 config the Dr. had an it said about 4000 hours for 100 images.
V96gb ram
Wit's true
VI had to edit his lora config as well, was too slow.
WI train an rank 128 lora fp16, I need 4 a6000 with 48gb Vram and nearly 5 hours
Wbut I get the best quality lora I ever have
VI can only do 64 rank since I need to work as well.
V2.61 s/it
Wthat why we use runpod or mass compute
Wbecause we can get something like H100 or A100 which speed is double or quadruple of the 4090
VYeah I plan on switching but wanted to learn how to do everything before being charged $$$ for failures.
DOn my 4090 it took 2.5 hours to train a lora
Wwhen I use 4 a6000 basically I have a vga with 4x speed of a 3090 and 48gb Vram
V@daroth How many epochs?
Wcost me only 4 usd for a lora
