https://civitai.com/models/328431?modelVersionId=573321
So what is good about this update
I have started training my LoRA at 1280x1280 instead of 1024x1024, this means higher res and easier details like hair clip and outfit
I had the old lora having flip augments on, this mirror'd her details sometimes and was just bad
other setting improvements
Get genning!
most subtle dramatard
Jump in the discussion.
No email address required.
Oh nice.
Do you think you could do like a beach/bikini one or is her outfit more or less baked into the lora?
Jump in the discussion.
No email address required.
it will probably work with a bit of tweaking, you might need to do things like put overalls into the negatives
Jump in the discussion.
No email address required.
That's likely a sign you've under tagged or over trained.
Jump in the discussion.
No email address required.
the model has so few images it's pretty much always going to be over trained , doing fewer than 2k steps also leads to garbage
Jump in the discussion.
No email address required.
Are you using synthetic data yet? That's the 6 month old fad for AI. Generate results, tag them good or bad and resubmit. Like a manual GAN.
Jump in the discussion.
No email address required.
I have long been using gens as data sets because there is only like 2 hand drawn images of her lol.
I just use my better cleaned up images
Jump in the discussion.
No email address required.
More options
Context
More options
Context
More options
Context
More options
Context
More options
Context
More options
Context