So you're saying that the stdout logs show 2100, but in the end the real count will be 8400?
So you're saying that the stdout logs show 2100, but in the end the real count will be 8400?
Sdr Furkan, exist controlnet canny or lineart for flux version that can be used in forgue?
Fyes batch size will be 4
Feffective batch size
Fi think i saw some early ones but probably not good. also i dont know if works with forge
E@Furkan Gözükara SECourses Hello Doctor, I am having problems with the presets tier 1 V2 slow and fast.
I seem to have a memory leak. The VRAM fills up, and then it eats almost all the RAM, the training becomes extremely slow and my PC loses stability so I have to urgently abort the process.
Do you know if Onetrainer updated any parameters? (I just executed the update.bat today) With the other presets or older versions of your own presets don't give any problems. Thanks.
I seem to have a memory leak. The VRAM fills up, and then it eats almost all the RAM, the training becomes extremely slow and my PC loses stability so I have to urgently abort the process.
Do you know if Onetrainer updated any parameters? (I just executed the update.bat today) With the other presets or older versions of your own presets don't give any problems. Thanks.
Dtier 1 is too much for any consumer GPU I think, tier 3 is the best that fits into ANY
MIs there a @Furkan Gözükara SECourses in the house? I couldn't find any material (post or video) about the importance of captioning and regularization images for SDXL Lora/Dreambooth training. Can you point me in the right direction?
MI got very good results with Flux (where you mentioned reg images are not useful, and a single trigger word is good for single persons), but my SDXL results where horrible…
Fyou probably loaded into dreambooth tab
Fand corrupted config
Falso hopefully i will update all configs today
Fit reduces likeliness
Fi tested it for person
FMedium
Check all the images below to see used training dataset and used captionings.
Fye tier 1 is for only 40 GB and above
Eoooh, for 40 GB of VRAM?
Fyep
Fgo tier 3
Frank 3
Fyou do flux right?
Ethat's explain a lot... hehe, okey, thanks! no, Trying to train SDXL
Ethe configs are only for flux?
Fi have different configs for each
Ffor sdxl use onetrainer it is bestr
Fyou using kohya or onetrainer?
Fi will link you accurate file
Eonetrainer
Fkk
Egotcha, thanks
FPatreon
Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon
Fuse tier1_15.4GB_fast_v4.json
Fthis is for sdxl
Fuse reg imags hugely improves for training a person
MThere it is, that was my second question. So you actually do prefer OneTrainer, at least for SD* models…
Fsd trainer really does handle vram better
Falso it is all in one
Fwith kohya we have gui and scripts
FHello together,
have anyone trained multiple concepts in one or more dreambooth training(s)?
I want to train 2 different characters with unique tokens into my model and face the issue that the one concept is always bleeding into the other.
I tried to remove the class token and made several tests with a combined training set and single trainings. Unfortunately no success. Each character has a versatile dataset of 35 high quality images.
Any idea to reduce the bleeding?
have anyone trained multiple concepts in one or more dreambooth training(s)?
I want to train 2 different characters with unique tokens into my model and face the issue that the one concept is always bleeding into the other.
I tried to remove the class token and made several tests with a combined training set and single trainings. Unfortunately no success. Each character has a versatile dataset of 35 high quality images.
Any idea to reduce the bleeding?
Fi just replied on youtube but it may not be possible since FLUX has internal text encoding. so it knows both are man. however you can try this. only ohwx and bbuk as rare token and use clip text encoder training that i added yesterday
Fstill may bleed. fine tuning may solve though not sure
FSorry for the missing information. I talk about a SDXL Model 
Fyes people said they do very good. are you doing lora or fine tuning?
FI fine tune the model with character images.
Lora training works great with your guides.
Lora training works great with your guides.
Fare you using onetrainer?
FI used OneTrainer (with your config) and dreambooth from kohya_ss also with your configuration
Fi see you tested on both
Fand still bleed?
FYes, onetrainer was much worse than kohya for some reason.
I made also dataset tagging with following schemas:
-> <unique identifier>
-> <unique identifier > <class token>
-> <unique identifier>, <describing image without character attitudes >
I made also dataset tagging with following schemas:
-> <unique identifier>
-> <unique identifier > <class token>
-> <unique identifier>, <describing image without character attitudes >
FTo test if the tagging is influencing common model tokens like men or women