Training A Public FLUX Style LoRA - All Process Will Be Shared
Added 2024-09-02 23:48:09 +0000 UTCI started training a public LoRA style (2 seperate training each on 4x A6000).
Experimenting captions vs non-captions. So we will see which yields best results for style training on FLUX.
Generated captions with multi-GPU batch Joycaption app.
I am showing 5 examples of what Joycaption generates on FLUX dev. Left images are the original style images from the dataset.
I used my multi-GPU Joycaption APP (used 8x A6000 for ultra fast captioning) : https://www.patreon.com/posts/110613301
I used my Gradio batch caption editor to edit some words and add activation token as ohwx 3d render : https://www.patreon.com/posts/108992085
The no caption dataset uses only ohwx 3d render as caption
I am using my newest 4x_GPU_Rank_1_SLOW_Better_Quality.json on 4X A6000 GPU and train 500 epochs - 114 images : https://www.patreon.com/posts/110879657
Total step count is being 500 * 114 / 4 (4x GPU - batch size 1) = 14250
Taking 37 hours currently if I don't terminate early
Will save a checkpoint once every 25 epochs
Full Windows Kohya LoRA training tutorial : https://youtu.be/nySGu12Y05k
Full cloud tutorial I am still editing
Hopefully will share trained LoRA on Hugging Face and CivitAI along with full dataset including captions.
I got permission to share dataset but can't be used commercially.
Also I will hopefully share full workflow in the CivitAI and Hugging Face LoRA pages.
Comments
yes it may get faster i constantly see updates
Furkan Gözükara
2024-09-05 09:48:09 +0000 UTCDo you think that the training process on this stuff will ever get faster? I thought that SDXL was insane compared to training SD1.5 but then it got faster. Now, I absolutely tolerate training an SDXL, although I know it won't do what FLUX does.
Diggy Dre
2024-09-05 07:59:46 +0000 UTCrecorded a fully tutorial and still editing sorry it is taking too long
Furkan Gözükara
2024-09-03 10:46:19 +0000 UTCyes it works. also recorded a fully tutorial and still editing sorry it is taking too long
Furkan Gözükara
2024-09-03 10:46:14 +0000 UTCfound the tutorial inside the file, will try it
Arvin Flores
2024-09-03 05:34:33 +0000 UTCany runpod tutorial? if not. Can i just git clone the flux branch? and run it in the terminal and http it so i can use it?
Arvin Flores
2024-09-03 03:01:58 +0000 UTCyou are welcome. i plan to search item training later like shoes
Furkan Gözükara
2024-09-03 01:46:23 +0000 UTCAwesome, I've been wanting to ask if you would do something that wasn't a person or that same data set.
Steve
2024-09-03 01:39:29 +0000 UTCyou can use this tutorial to train yourself and you can generate yourself as playing a golf : https://youtu.be/nySGu12Y05k
Furkan Gözükara
2024-09-03 01:31:54 +0000 UTCtrain a lora for palying golf
shen oracle
2024-09-03 00:57:42 +0000 UTC训练一个打高尔夫球的lora
shen oracle
2024-09-03 00:57:07 +0000 UTCcan you explain more in details?
Furkan Gözükara
2024-09-03 00:54:13 +0000 UTC来和姿势训练吧
shen oracle
2024-09-03 00:51:58 +0000 UTC
