wow ipndm makes it much more realistic
wow ipndm makes it much more realistic
D@Dr. Furkan Gözükara in my case instance Prompt is the keyword trigger?
but in class prompt what do I write? it is not a person
but in class prompt what do I write? it is not a person
DI will try a first test using:
Rank_1_FAST_4x_GPU
Rank_1_FAST_4x_GPU
Fuse joy captions and like chatgpt
Fto get your images described
Ffrom there get idea
Flike i used 3d rendering
Frank 1 fast is good
Flol i publish thousands of dollars worth article, lots of research info and still some people complains like why i use lora rank 128 instead of 32 
you can use if you wish
you can use if you wish DAccording to chatgpt it tells me that my image is a “Countryball”.
but using some caption models trained in replicate or huggingface it tells me it's a “cartoon illustration”.
but using some caption models trained in replicate or huggingface it tells me it's a “cartoon illustration”.
Ftry both on flux
D
Fand see which is more accurate generating
Fcartoon illustration looks more accurate
Dis it necessary to describe that it is in 2D?
DI have a doubt, and excuse my ignorance....
but when using the prompt is it necessary to write “cbznft cartoon illustration”? i.e. use both?
or am I wrong?
but when using the prompt is it necessary to write “cbznft cartoon illustration”? i.e. use both?
or am I wrong?
Fye like this
Lstupid question: should or can training images also contain things to learn which are rare? like 98% of all the training images of the same person have blue eyes -- and 2% have green eyes (like they wearing colored contact lenses)?
Lanother example -- how the person looks like from "behind"
Fthey should contain images
Fonly you want as output
Fafter training
Fi feel like this new sampler beter
D@Dr. Furkan Gözükara I am going to try without *.txt captions, 800 images 1024x1024, what epoch should I use? I think there is a formula, and how often the checkpoint?
DSave every N epochs?
D
Dthat's the tricky part, understanding how much to use epoch, train batch size. 
Ffor 800 images
Fsave every 1 epoch
Ftrain batch size max 2 brings some speed up
Fon rtx a6000
Depoch?
Fhard to tell
Fi feel like max 50 should be sufficient maybe much lesser
DI will make a first test with 25
Dletsgo!
Dit is normal?
Dto infinity and beyond. Poor GPUs must be on fire.
Fyep looking good
Fposted a new comparison : https://www.reddit.com/r/FluxAI/comments/1fbnfl3/sampler_unipc_unified_predictorcorrector_vs_ipndm/
Reddit
Explore this post and more from the FluxAI community
DGPU 2, 90 C* I think it needs maintenance
Djajaja
FFeven 100 is good
DI agree, it adds a bit of grain to the images. It is also very acceptable for multiple loras. I have used like 5 with it together, and it worked marvellously.
DThe grain makes images seem more natural, realistic when viewed without zooming in. Zooming in is a different story.
DIssues I've experienced with the new sampler: SUPIR upscale removes the grain from its output and smoothens it out again.
Fare you using latest default configs?
Fit is very loyal
Dis it possible after training my lora, to test it with the flux shnell version?
DTraining completed using 2 GPUs on Mass Compute. The experience was great, much better than Runpod, I may never go back to Runpod again after this.