thank youDr Furkan, do you mean the half model lora in kohya?
thank youDr Furkan, do you mean the half model lora in kohya?
Fit will train half of the weights of the model
Fto fit into 16 gb
Fbut maybe more optimizations come
Fi am following
Fso it will be reduced quality training
Fi am making installer for face fusion next branch right now
Aany of you guys know how to fix this? on comfyui with on the face detailer section in scheduler
Muse spandrel
Mthe one youre using is deprecated
Maparently
MGitHub
Spandrel gives your project support for various PyTorch architectures meant for AI Super-Resolution, restoration, and inpainting. Based on the model support implemented in chaiNNer. - chaiNNer-org/...
Fnot that i know but i use swarmui segment
Fworks amazing
FPatreon
Get more from SECourses: Tutorials, Guides, Resources, Training, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözükara on Patreon
V@Dr. Furkan Gözükara can kaggle be used to train 1.5 model now?
WI am using the updated SwarmUI. I installed the new nf4 bnb stuff and told it to install bitsandbytes but now get this error when trying to generate with that model:
Wmessage.txt2.59KB
Fdid you restart?
Falso nf4 models goes into stable diffusion folder not unet
Fyes it can be used
Fwe have kohya config too
Fand notebook
Vokay also 1.5 ckpt training, right? i have to retrain models
Fgive it a try notebook on patreon
Vokay let me go over it now, thanks a lot
Wyes it's in the stable diffusion folder and I did restart.
Fplease ask in developer discord he answers all
F
O(Updated both versions, both 0.1 and ComfyConverted are the same file now. please redownload.) Deep Semantic Comprehension: Flux+ Detailer excels i...
Awhat is this? and how doi fix it?
Danyone else wondering if SD has a reaction to flux?
TFlux only works good with euler simple?
RIt also works with deis sampler and with the beta scheduler - I didn’t see much difference between them though
Fthey said gonna publish sd 3.1
Fi am waiting
Fi havent tested yet i should test
HSd3.1 will still only be 8b vs flux at 12b right?
Fno ideas
Fi also dont expect much
Fflux is many levels better than anything else
TIn MC there is some kind of error with Forge, when I put the arguments export COMMANDLINE_ARGS="--xformers --share --enable-insecure-extension-access" there is no way to start it, the console closes by itself. I think the problem is with xformers.
Tyes, the problem is with xformers 100%, with the other arguments work
Fye with latest version xformers not working
Ffor everyone
Ftesting 2 new configs for onetrainer
Fdoing 4 th training
F3 with one trainer 1 with kohya for sdxl
Fperfecting it before moving to flux :d
Vwhen will you create LORA training for flux? and did you try the latest flux nf2 v2 model? and do you think flux is totally going to replace stable diffusion? sd 1.5 and sdxl...