Ver. 1 . If you like my work and want to support me you can also do support me by donating Buzz or by PayPal https://www.paypal.com/donate/?hosted_button_id=L87JAGTHDNBRA
Note: Please if you see any of my models somewhere else report or let me know! Thnx
Support creators:
by liking their models (many forget to do it)
by publishing your photos directly in the model's gallery.
by getting involved in their Early Access programs
Thank you for your support.
Description
Ver.1
FAQ
Comments (8)
It does something, but can't get the shape/look from your example images, and varies depending on checkpoint. RAW+turbo lora is the worst and I feel it doesn't do much (I tried with the textfusion refusal lora alone and with fedor filter bypass), while a couple other ckpt just give puffy nipples (nail bat example prompt), none look close to your examples.
remember not all CKP are trained the same or merged the same so the lora also be affected mostly from the main model. Also here in Civitai in my experience at least the generator is terrible a lot of the time. But if your using it locally try different samplers and schedulers I do it all the time. And Remember Turbo is just distilled (degraded) in other words even if a turbo lora is compatible with a full models the effect wont be the same. By the way did you tried raising up the lora weight?
what text encoder are you using?
@Vermodia huihui Abliterated 4b instruct.
@Artikuz_Ai I'm running locally. I'll keep messing with it, ckpts, prompts, schedulurs/samplers. Just tried increasing strength passed 1.0 and it's helping. I was worried of it getting overfitted and showing the dataset.
@AI_DK good, so what model your using? well the dataset was made with very selected i mean specific real banana shape to get that long curve breast so unless the user decide less effect and adding more that +1 i haven't try but may be exaggerated or overcook a bit.
@AI_DK @Vermodia I also use Abliterated 4b instruct is pretty good
I mean, you may wanna remove the 'realistic' part of the name ;)