OneTrainer
OneTrainer copied to clipboard
Question about Latent Caching and Concept "Image Variations" setting.
What happened?
I am about to finetune with 75 photos with the latent caching setting on.
I read the wiki and it says to not use the default "1" image variation setting when using latent caching.
Though I still don't really know what to set the "Image Variations" setting to under Concept then.
Any ideas? My best guess could mean if there's 4 images of myself in there from the same scene (different camera angle) thats 1 variation? I counted this way and out of the 75 there would be 50 variations then, I have mostly different photos. Not sure if that's actually what this means..
This is what the Wiki says currently:
Latent caching Latent caching (as described in the caching section) will speed up the training by saving intermediate data. If this setting is enabled, some data will be calculated based on your training images and saved to disc. If you enable data augmentation, you should increase the "Image Variations" setting of that concept, otherwise only a single augmentation version will be cached.
What did you expect would happen?
The Wiki covering "image variations" entry based on how many images you are using, if latent caching turned on.
Edit: I also noticed the "text variations" setting. I have 75 text files captioned. Would I put 75 as the setting? I would prefer to use a single text file in the concept but not sure the format it it requires, or how to specify separate token and class there. Edit: nm on this part selected all files and caption.txt was hidden in the 75 images. Needs .txt drop down option to avoid confusion.