Can you share some images from your trainings comparing the normal dev vs de distilled one?
Can you share some images from your trainings comparing the normal dev vs de distilled one?
DNot from my training because is from family and friends I can't share, The quality on inference is identical, the advantage is that you can train multiple people on the same lora, I trained 11 people in one lora. I'm perfecting the technique to avoid concept bleeding completely not only between the people I'm training that is fixed, but fix the class bleeding to a minimum too.
�Okay thanks for all your updates here
Dyou welcome.
DMy training on de_distill just completed (dreambooth on a single person). The ability to control cfg helps you create a far greater image than CFG 1 @ flux_dev. The results are absolutely amazing. You can create more noisier (and less plasticky images) which seem more realistic. The ability to use negative prompts and the prompt adherence is very, very good.
Only downside is: Time! It sure takes helluva lot of time to generate (Minimum Steps 70 imo), and there are extra inference parameters to deal with.
Only downside is: Time! It sure takes helluva lot of time to generate (Minimum Steps 70 imo), and there are extra inference parameters to deal with.
DMy training on dev2pro with the same parameters (same concept) would also be completed soon.
Dgreat!!!
Ddev2pro was made to train loras and then use them on regular flux I think, is not really de-distilled
DI never try it, but I can try and see if it helps with concept bleeding.
DIt is not truely de-distilled true.
DI don't like what it does to hands so far lol
DWith thresholding, it is able to generate nicely at CFG 3.5.
DIt's in a strange place b/w flux dev and de-distill. 
NI still use jegernaught and which models in Super now give the best result for realism for CG graphics? 3D, 2D graphics and drawing for example anime
Fi didnt test for it actually
Fnow i need to record intro part of the tutorial
Ftime to edit
F
Aanyone know if its possible to increase the vocab of Flux Dev ? Like to train the clip models to add new vocab ?
DYes, you can train the clip or T5 through captions
Aboth at the same time ?
Fno way to increase as far as i know. everything still will get tokenized
AI think i figured it, I increased it but havent tested it yet
Fi turned off fast boot and hibernate
Fstill getting random BSOD
Flets see if this will fix
WHi @Dr. Furkan Gözükara what is the best way to check whether our lora or checkpoint is overfitted
Wwhether is not follow the prompt style or body proportion is not normal or something else ?
Fanalyzing grid
Fin the upcoming video i have explained
Fediting right now
FThis is why you should get minimum 64 GB RAM memory
WGreat Im looking for it
Fkohya is working on fixing bleed issue
FDifferential Output Preservation(仮)、FLUX.1で学習してみた。prior_loss_weightは10。一応効果ありそうだけど、diffがだいぶ学習不足なだけかもしれない……。
10/19/24, 10:10 AM
Mno we have to wait for bmaltais to merge to gui
Fyep
Fthe videos waiting me to edit :/
F
Mkohya updated
Mnow we wait for bmalt
SI'm looking for ComfyUI workflow with Flux Schnell. Anyone got ?
DSThanks!
SDoes anyone know a good tutorial to use runpod-worker-comfy ? I just want to deploy an API to generate images https://github.com/blib-la/runpod-worker-comfy
GitHub
ComfyUI as a serverless API on RunPod. Contribute to blib-la/runpod-worker-comfy development by creating an account on GitHub.
DUpdate https://huggingface.co/nyanko7/flux-dev-de-distill Lora training, removing the class did not help with class bleeding, the result was almost the same, and maybe worse because difficulted the prompting on inference. I'm doing another test with just one class and some regularization images, idk if kohya new code was implemented yet, hope this will help too.
PNew technique I'm introducing for identity preservation.
Fwow this is good
Amore information please
Pthis feature is part of new webui in development, qwen-flux-wan-such This feature is part of a new webui under development, when finished, Dr. will make a tutorial on how to use it. For now I'm in the final stages and presenting some previews here. This feature only uses the image-to-image diffusion pipeline and a controlnet to refine.
