Trigger word
None required (fully trigger-free) The LoRA activates purely from natural language descriptions.
Please see my example photos for an idea of what is working currently.
It seems to default to more Asian looking women unless you specify a look.
I'm working on a larger dataset to increase options as it feels limited in some areas but when you do get the results they are amazing! I love this new model and how to treats facial expressions compared to PonyXL ect.
Base model
Z-Image Turbo (all samples generated on pure Z-Image Turbo)
Generation info for all sample images
Z-Image Turbo, Euler, 9 steps, CFG 1.0, LoRA strength 0.4–1.2
UPDATE: 0.4 weight seems to be better with keeping face information and mixing with other loras.
What this LoRA does
Trained on a hand picked dataset of the most photorealistic tentacle scenes I could find. Extreme cumflation, deep oral/anal/vaginal, tight wrapping, suction cups, ridges, slime, post-climax overflow).
Recommended Settings
Base: Z-Image Turbo Sampler: Euler Steps: 9 CFG: 1.0 LoRA strength: 0.4 – 1.2 , lower the weight if using with a character lora.
☕ Support
Enjoying my Loras? or want custom commissions? [Buy me a coffee]
Help fund more training using my runpod referral link, We both get credit! [Runpod ]
Description
Improvements to working with Character LoRAs. Tentacles more consistent, black large tentacles ''cumflation'' is working as intended. It was a bit hit and miss in v0.8. I'm happy with my dataset now and looking forward to training with z-image base when released! Works with lower weights 0.7-1 when using with a character lora.
FAQ
Comments (8)
It works, but it blows away a character lora's face and other trained details that a tentacle might overlap or interact with.
EDIT: At the creator's suggestion I swapped out rgthree's PowerLoraLoader with the native LoraLoaderModelOnly and it fixed the clobbering. Really good lora!
What weights did you use on your LoRAs? I find that 0.6–0.7 for this model works pretty well, and then I push my character LoRA slightly above 1. Also, make sure you’re loading the character LoRA with a LoadModelOnly LoRA loader and not stacking it. Stacking seems to cause a lot of issues. The workflows linked to the example images are actually bad. Don’t use those, as they were causing problems for me.
I tried 0.8 down to 0.3, but i didn't try turning up my character lora. Also didn't try modelonly. Is the rgthree power lora loader really an issue?
Ok, you were right. I tried it with LoraLoaderModelOnly and it doesn't clobber the face. It still has a minor effect on it, but not so much you notice given the kind of lighting and ambience you'd want in a scene with this model. Thanks for replying and calling that out, much appreciated!
@fantastorium No problem! I will be training it again once z-image base is released and hopefully fix some issues.
@fantastorium Interesting. I wonder if pythongosssss/ComfyUI-Custom-Scripts Lora Loader is okay to use or not.
@fantastorium The only difference is the Power Lora Loader takes CLIP in too. Can you just test what power loader does if you don't include CLIP?
dunno bro, to me the limit for this kind of thing anime style