you can convert model to fp8
you can convert model to fp8
FPatreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
Fdont caption in details
Fit reduces likeliness
Fand you already dont have many images
Jthank, I will try with more images about 40 images, should I keep epoch 200 or lower epoch?
F200 good
Fso you can be sure it is trained enough
SI created a pod from a network volume and now want to store files in the volume
Swhere do I find the network volume inside the pod when I'm ssh'd in ?
Fyou can ask this in runpod server i never done
FFi just found a new error of bmaltais
Fnever ending :d
FGitHub
Hugging face hub has to be 0.25.2 or doesnt work man these errors always happens to me when recording a tutorial video @kohya-ss @bmaltais
Ti would disagree partially. u wanna caption clothing, background etc.
but not for what the lora should learn from the images
but not for what the lora should learn from the images
Fwell flux has internal captioning mechanism
Fthat is why i dont recommned for flux
Tit's just.... i trained same 10img lora of myself once without captions and once with minimal captions. and it's so hard to say which is better side by side... subjective.
i need some more xy grids
i need some more xy grids
Tdid BFL tell you? ^^
Fwell it is discovered by several scripts developers
Fby analyzing its architecture
Fit generates image embeddings
Fwhich makes captioning effect
Tinteresting. thanks for sharing
P@Dr. Furkan Gözükara Hello everyone, I have a question about custom lora training on kohya using flux
I trained my face with 21 photos and 142 epochs as you mention on the equation on your video, which is... 3.000 / (number of photos) = number of epochs.
When I write the prompt on SwarmUI using Flux.1-dev-fp8. with 1024x1024 resolution the picture looks with low quality, also when the size is 1024+ seems to appear some artifact vertical lines on the picture (see the image) . I'm thinking to train again from zero a new lora with more images or use finetuning. What do you recommend?
I trained my face with 21 photos and 142 epochs as you mention on the equation on your video, which is... 3.000 / (number of photos) = number of epochs.
When I write the prompt on SwarmUI using Flux.1-dev-fp8. with 1024x1024 resolution the picture looks with low quality, also when the size is 1024+ seems to appear some artifact vertical lines on the picture (see the image) . I'm thinking to train again from zero a new lora with more images or use finetuning. What do you recommend?
Ffine tuning prevents this
Fbut this usually happens when you generate high resolution
Fah yes you said 1024+
Fyes fine tuning reduces this a lot
Falso you can use SUPIR to upscale
Fsupir face enhance difference
F
Foriginal vs supir https://imgsli.com/MzA4NzQx/0/2
Dis there a recommendation for a good training set? (background, composition....)
Fdifferent background and clothing
Fwill yield better results
Fso model wont memorize
Ffine tuning tutorial this will take forever to edit :/
IHi everyone, I'm not sure which chat to post this in, but can someone explain to me how to properly use wildcards in swarm ui? I have a weird issue with. I have a wildcard file with 10 lines for shirt color in it but when I try to generate 10 images, it will choose one color from those 10 and generate 10 images with that color
Fswarm internalk
Fcheck those parameters
F
IHmmm I will look for it
IThank you
Nthat worked thanks
DUpdate https://huggingface.co/nyanko7/flux-dev-de-distill training. LR 0.0001 vs 0.00003, much better results with lower LR, may be a little undertrained. Less class bleeding, but the class on training captions seems to decrease resemblance in non de-destilled models, I think that adding the class is a problem, I will remove the class from the captions I think that it will eliminate class bleeding completely. You can always add the class on inference if you wish, but removing it from the captions will protect the class from bleeding. flux-dev-de-distill learns very different than regular flux-dev it learns the caption tokens much better. I will update tomorrow-
�Can you share some images from your trainings comparing the normal dev vs de distilled one?
DNot from my training because is from family and friends I can't share, The quality on inference is identical, the advantage is that you can train multiple people on the same lora, I trained 11 people in one lora. I'm perfecting the technique to avoid concept bleeding completely not only between the people I'm training that is fixed, but fix the class bleeding to a minimum too.
�Okay thanks for all your updates here
