Why don't you try simpletunner?
Why don't you try simpletunner?
AI've been using config files from KOhya_FLUX_DreamBooth_v17, after installing the latest version of kohya_ss for the sd3_flux branch, and my Flux dreambooth sessions on an RTX4090, which used to sit at about 23.9 GB in VRAM, and they'd run fine, just don't sit within the VRAM 24GB size anymore. I've been fiddling with block_swap size, but 30.7 GB seems to be the best I can get for memory usage with at least some iterations happening. Is this a known thing? It could be that I've set some parameter somewhere when setting up kohya_ss again.
AI even tried rerunning a previous .toml script for dreambooth, on the command line, but it didn't repeat the previous VRAM experience of using under 24GB.
AIt isn't super-slow, by the way, running at 5.00 - 7.00 sec/it.
CHi everyone!
@Furkan Gözükara SECourses can you give me a hint?
I have this error:
RuntimeError: CUDA error: out of memory
"CUDA kernel errors may be reported asynchronously by some other API call, so the stack trace below may not be correct.
For debugging, consider passing CUDA_LAUNCH_BLOCKING=1"
I tried 8gb and 3080 10gb config
I have 48gb RAM and 10gb VRAM.
What can help me?
@Furkan Gözükara SECourses can you give me a hint?
I have this error:
RuntimeError: CUDA error: out of memory
"CUDA kernel errors may be reported asynchronously by some other API call, so the stack trace below may not be correct.
For debugging, consider passing CUDA_LAUNCH_BLOCKING=1"
I tried 8gb and 3080 10gb config
I have 48gb RAM and 10gb VRAM.
What can help me?
C@Furkan Gözükara SECourses Hey. I saw your yt community post, and wondering will there be a tutorial coming for sd3.5? maybe on patreon post early.
Fnope all data lost once you convert into fp8
Fi dont plan to invest time into that one
Fthis can happen
Fif you loaded lora into dreambooth
For dreambooth into lora
F5 second it is good
Ffor flux dreambooth
�Way better than ai toolkit btw
Fhi when you have it?
Fyou are out of VRAM
Fi dont like the developer, the guy attacked me so many times without any reason 
Fcould be currently i am doing research for a client
�Yes he is rude sometimes but a great developer and he very well documented everything and very active in discord servers.
Cnice looking forward to this, since flux have its own disadvantages, im still using sdxl. hoping sd3.5 will amaze me.
Ccan we use regularization imgs with sd3.5?
Fi think people said same as flux but havent tested yet
Fwill test
C
Clooking forward to this
Clooking forward to this
SWould it be possible for you to perhaps share any results? I am also looking into simpletuner and would love to see if someone has achieved some reasonable results. Would really appreciate it, I won't share ^^
�I have not yet tested it but I plan to do that next month. I have heard many people talk good about it. Here is a video that goes deep into it and compares it with ai toolkit - https://youtu.be/se3qpLkJnrk?si=7MhyT8i3DALGK9L5
YouTubekasukanra
#flux #simpletuner #bghira #ComfyUI #LoRA #finetune #stablediffusion #style #styletraining
This is an FLUX.1 [dev] LoRA training log for art style and concepts. I document my thought process, experiments, mistakes, and analysis of quantitative and qualitative results. Hopefully, this can be a good starting guide for people looking to train LoRA...
This is an FLUX.1 [dev] LoRA training log for art style and concepts. I document my thought process, experiments, mistakes, and analysis of quantitative and qualitative results. Hopefully, this can be a good starting guide for people looking to train LoRA...
SSo your opinion "Way better than ai toolkit btw" is based on a video? You haven't tried it yourself yet?
SInteresting
�See the video you will know. I have done a lot of trainings with ai toolkit btw
Ncan some explain how to reduce flux lora size with no (almost) quality loss? Because I've made bunch of them and 1gb loras is eating my storage) or maybe there is video with explanation?
Nnever mind, found in patreon posts
Fsaving as fp8
Nyep, reduced to 260mb instead of 1GB, thx
Fyou are welcome
GHi @Furkan Gözükara SECourses , do you have an idea how to create a consistent character for image generation from a single image?
Fyes 2 step training
F1st traing with single image
Fadd cropped version zoom out etc
Fthen generate images
Fthen train again
GInteresting idea, I'll try it. Thanks
GI saw this tutorial, I wonder if they do it the same way behind the scenes
https://youtu.be/UY_aRiOK6gY?si=6U4TAXFPa0lxPPU0
https://youtu.be/UY_aRiOK6gY?si=6U4TAXFPa0lxPPU0
GWhat about this process using different posing on a single image?
https://youtu.be/MbQv8zoNEfY?si=oufZgsSpJJSBMGD9
https://youtu.be/MbQv8zoNEfY?si=oufZgsSpJJSBMGD9
Fsame logic. if you can get more poses from single image can help
Fbut still i recommend twice training
ASingle Image training is pretty good but will work exponentially better if you can get at least three from different angles from my experience
Ayou will save yourself a lot of time and headache
AIf you have Comfy there are some workflows floating around which will attempt to create these angles from the image - you can randomly hit on some with a few rolls of the dice
Lunder this logic, not having so many images, could you recommend some configuration to train with a minimum of 5 or 10 images, how many steps, etc.