Is the Expose HTTP Port correct?: 8888,3001
Is the Expose HTTP Port correct?: 8888,3001
Ddid you post some of your prompts anyware to test myself?
Fyep
Fprompts included in patreon zip file let me update to latest prompts
Fi will also make a new post
Fmore tidy
FPatreon
Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon
Dthanks!!!!!
Sgod bless you Dr Furkan
Dwhat is the ideal max new tokens value to prevent hallucinating? Also I installed V9 in windows, Is there any way I can simply upgrade it to V10 without re-installing? (Joycaption)
Dwhat is the prompt input instruction being used in Joycaption by default? Can I customize it?
Fjust overwrite app.py file
F300 tokens i think good
Fno instructions it is fine tuned
Fi am working on multi gpu support
DWhen is Supir new local version launching (V2), any dates?
Fnope not yet
Fman multi gpu app is hard
Ftesting multi gpu joy caption
Fgonna caption 20k images :d
WQuestion. Is it possible to train a Lora on 2-3 images of a person if that’s all I have? Also how many should be used to train a style Lora?
Fas many as possible for style
Ffor sdxl 200 was decent
F2-3 images may work but i didnt test
Fusually you need to have different poses distances and expressions if you need
FI'm trying to get SwarmUI working and it's not being very cooperative
Fmake cfg 1
Ffor flux
Fi made joycaption multi threaded
Fit is slower than 1 thread but 8 gpu works great
FTy, now the result looks correct
F
FTesting another LoRA
DHello all, if train of 8 photos on two different model will the training fail? or will it come with an average of the two faces?
I'm asking because i trying to mix two models as to come as with a new likeness but the training refuse to finish
I'm asking because i trying to mix two models as to come as with a new likeness but the training refuse to finish
FWas trained on images of this character
Fit may get mixed bleed
Dyeah i'm looking for that, i want a new person that is neither
Das long as each output is the new one
Di'm happy
Dbecause i wanna create a virtual influencer that ressemble neither of them two
DBut instead each output give me new one
Ftry it
DIs the fune fill tuning course up for sale?
Fit is shared here
FPatreon
Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon
Fit keeps gettinbg updated
Fonly for lora
Ffine tuning next hopefully
Fyep it is super faithful but it overfit as well. i think fine tuning will be way better than lora
ADr how big are your safetensors from flux trains? Mine are 300-400mb instead of 6gb files lol
Fyou train lora?