A fun compact and portable lora, trained at about 3000 steps, 47 images, rank 8, at 512x512
Works best with euler \ beta, use whatever feels best for your case; min 18-23 steps, start at 0.8 depending on the model, but 1.0 is the standard
AVOID any turbo loras unless you'refine with loosing some quality
ALWAYS be descriptive of what you want, an easy test is to go on twitter, find a random fursuiters account, click a photo, and just type out exactly what you see,
If you're not going to be descriptive, then just keep it straight to the point.
and ALWAYS include the trigger, unless you want a softer tone to mix with other loras; and it does actually work pretty well with other loras
Description
version 1
FAQ
Comments (3)
The sample image is the best I've ever seen.
But I don't know why it doesn't work well in my hand.
I'm not even sure if it was activated or not.
Can you show me some samples of ComfyUI Workflower JSON with the model and value u use?
There are too many settings that will make a big difference in anima.
after testing it with 4 different photorealism models:
panima_v1.0
photoanima_v2.3turbo
samANIMARealistic_turbov2.1
samAnimaRealistic_turbo2.3,
the only two parameters to look at are CFG and steps, if its a turbo model, keep it around 12-14 and 1.0-1.7, if its distilled, just use a turbo lora and make it about 14 steps,
ive noticed i get the blurry photos when using either a too low cfg or using a turbo lora on an already turbo model, outside of that, just ensure that your CFG is correct (like panima prefers lower cfg, sam prefers higher) and set your steps around 12, everything else im using it standard, euler + beta, random seed, only the lora + model, nothing else lol
@daspin335 Thank you so much, I'll try it!