Trained Myself With 256 Images on FLUX - Results Mind Blowing
Added 2024-09-14 22:37:09 +0000 UTCDetailed Full Workflow
Medium article : https://medium.com/@furkangozukara/ultimate-flux-lora-training-tutorial-windows-and-cloud-deployment-abb72f21cbf8
Windows main tutorial : https://youtu.be/nySGu12Y05k
Cloud tutorial for GPU poor or scaling : https://youtu.be/-uhL2nW7Ddw
Full detailed results and conclusions : https://www.patreon.com/posts/111891669
Full config files and details to train : https://www.patreon.com/posts/110879657
SUPIR Upscaling (default settings are now perfect) : https://youtu.be/OYxVEvDf284
Details
I used my Poco X6 Camera phone and solo taken images
My dataset is far from being ready, thus I have used so many repeating and almost same images, but this was rather experimental
Hopefully I will continue taking more shots and improve dataset and reduce size in future
I trained Clip-L and T5-XXL Text Encoders as well
In the above shared images the 19th image is the used dataset, 256 images, and 20th image is the comparison with 15 images training dataset and several checkpoints of newest training
Since there was too much push from community that my workflow won't work with expressions, I had to take a break from research and use whatever I have
I used my own researched workflow for training with Kohya GUI and also my own self developed SUPIR app batch upscaling with face upscaling and auto LLaVA captioning improvement
Download images to see them in full size, the last provided grid is 50% downscaled
Workflow
Gather a dataset that has expressions and perspectives that you like after training, this is crucial, whatever you add, it can generate perfect
Follow one of the LoRA training tutorials / guides
After training your LoRA, use your favorite UI to generate images
I prefer SwarmUI and here used prompts (you can add specific expressions to prompts) including face inpainting : https://gist.github.com/FurkanGozukara/ce72861e52806c5ea4e8b9c7f4409672
After generating images, use SUPIR to upscale 2x with maximum resemblance
Short Conclusions
Using 256 images certainly caused more overfitting than necessary
I had to make prompts more detailed about background / environment to reduce impact of overfit, used Claude 3.5 (like ChatGPT)
Still FLUX handled this massive overfit dataset excellently
It learnt my body shape perfectly as well (muscular + some extra fat)
It even learnt even my broken teeth or my forehead veins perfectly
The outputs are much more lively and realistic and has better anatomy
I couldn't get such quality photo in a professional studio as in image 18 - the quality and details next level
Since dataset was collected at different days, weeks, months, the hair, the weight of me, the skin color was not consistent, which caused some different hair style and length or skin color at inference :D
Comments
email me entire CMD logs this doesnt tell error : monstermmorpg@gmail.com
Furkan Gözükara
2024-09-15 22:19:17 +0000 UTCCan't get anywhere with this. Seems as though your installer didn't install everything correctly: Traceback (most recent call last): File "C:\Users\andre\AppData\Local\Programs\Python\Python310\lib\runpy.py", line 196, in _run_module_as_main return _run_code(code, main_globals, None, File "C:\Users\andre\AppData\Local\Programs\Python\Python310\lib\runpy.py", line 86, in _run_code exec(code, run_globals) File "J:\Kohya_GUI_Flux_Installer_v33\kohya_ss\venv\Scripts\accelerate.EXE\__main__.py", line 7, in File "J:\Kohya_GUI_Flux_Installer_v33\kohya_ss\venv\lib\site-packages\accelerate\commands\accelerate_cli.py", line 48, in main args.func(args) File "J:\Kohya_GUI_Flux_Installer_v33\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py", line 1106, in launch_command simple_launcher(args) File "J:\Kohya_GUI_Flux_Installer_v33\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py", line 704, in simple_launcher raise subprocess.CalledProcessError(returncode=process.returncode, cmd=cmd) subprocess.CalledProcessError: Command '['J:\\Kohya_GUI_Flux_Installer_v33\\kohya_ss\\venv\\Scripts\\python.exe', 'J:/Kohya_GUI_Flux_Installer_v33/kohya_ss/sd-scripts/flux_train_network.py', '--config_file', 'J:/Kohya_GUI_Flux_Installer_v33/Training\\model/config_lora-20240915-163812.toml']' returned non-zero exit status 3221225477. 17:17:54-386811 INFO Training has ended.
JustDrwIt
2024-09-15 21:52:08 +0000 UTCyes we save as float, doubles size, we train lora rank 128 increases size, training clip l also increases size
Furkan Gözükara
2024-09-15 16:43:13 +0000 UTCHi, thanks for all of this. When I use CivitAI's flux lora maker i get a lora which is around 200mb. When i used your Rank 2 - each lora is 2.5gb, is there a reason for that?
Erosi
2024-09-15 07:08:44 +0000 UTCT5 minimally improves maybe it is just placebo but clip l really improves I had shared grids
Furkan Gözükara
2024-09-15 03:51:25 +0000 UTCThanks for your finding ! Could you share more information about "Train Clip-L and T5-XXL Text Encoders"? How does it affect the result?
Sugar Coat VFX Design
2024-09-15 03:45:12 +0000 UTCYes hopefully it is research of this week
Furkan Gözükara
2024-09-14 23:59:11 +0000 UTCcould u make a flux finetune fulx-dev
shen oracle
2024-09-14 23:54:14 +0000 UTCthank you so much
Furkan Gözükara
2024-09-14 23:47:42 +0000 UTCamazing work
shen oracle
2024-09-14 23:39:14 +0000 UTCAI may pick those AI made fake backgrounds and changes be careful :D therefore I suggest real images to be used unedited
Furkan Gözükara
2024-09-14 23:16:23 +0000 UTCI took mine with different lighting setups and some different expressions but mostly closer up headshots then I took into Photoshop and used generative fill to change the backgrounds and outfit on every image but I'm going to try out some more angles and lengths in. My first test I had a combination of phone selfies and DSLR shots at 85mm but I think the phone selfies because of lens distortion actually hurt my first test as soon as I removed them and used focal lengths from 50mm+ it helped keep my facial proportions more accurate
Brent Young
2024-09-14 23:14:47 +0000 UTCyes they were overkill. i think adding a unique photo from each desired perspective , emotion, expression, distance and angle is the best one instead of adding similar ones as i used in the dataset :D
Furkan Gözükara
2024-09-14 23:09:52 +0000 UTCSo would you say that many images was overkill? What do you think a great dataset size would be? I have done a couple on 20 images using DSLR images and my results have been fantastic but I'm always looking to improve it
Brent Young
2024-09-14 23:01:42 +0000 UTC
