VR Porn
My Fans Lora : fanslora.com/miaou
Trained with +5k HD Pictures
Use Trigger word and negatif prompt like me for good results
V3 : Trigger Word Avalable (some work better than others), better bad body result but you may still bad result sometime, be patient.
Recomanded Setting :
Not use : "Restore faces" if you upscale
Upscaler : Lanczos x1.3-1.5
CFG Scale : 4-7 (low value for difficule possiton)
Have fun and don't hesitate to share your images !
Credite if you merge this model.
Description
Bad body fixed but not perfect.
You can use normal prompts now.
FAQ
Comments (21)
Can this model do cunnilingus?
I think cunnilingus it's a bit complicated
there are loras
@phili13251 yeah, I know, but those don't always work correctly or it's difficult to get two full bodied characters in the shot. At least for me as I can't highresfix cause I only have 4vram
@Someone5544 hae u tried -medvram or --xformers ? (higher than 768x doesnt improve the understanding for the ai anyway)
@phili13251 I use medvram and It's interesting you say that higher than 768x doesn't help to improve it as I've had a friend make a lora for me with it and for them it could work and gen two full people in the same pic doing actions, however they said it NEEDS highresfix, and I'm not sure what that entails as to why I need it when others I don't. As in what in training a lora makes you need to use vs not needing it. Same goes for models it seems, some work great at 512 or 768 (highest I can go without OoM errors) yet others I've tried that have had great reviews and pics to back it up, I couldn't make anything work with it cause I couldn't use highresfix. I'm working on getting a better PC, but it just still strikes me odd how highresfix, at least sometimes completely changes the entire picture instead of just well, making it look more high res, lol
@Someone5544 a little late response, but have a look at "tiled VAE" and "tiled diffusion". i can do hires.fix at the full 4x upscale with those without running out of ram or vram.
A thought - if you don't want people to be confused and need them to follow a specific prompting method rather than give you bad reviews, make those prompts plain in the description rather than hiding them behind bit.ly. A LOT of anti-spam tools consider bit.ly a bad actor and won't follow the link, at which point you are telling people they need to follow instructions they then can't see - and then you're surprised you're getting bad reviews?
I can confirm that I consider bitly a bad actor site and is blocked. It wouldn't cause me to give a bad review, unless I can't get good images from the checkpoint.
People should not be giving reviews if they don't know how to press the little circled i button to get prompt info
You can also download the CivitAI Helper extension to see the full prompts for each image + be able to send them to txt2image directly if pressing the little round button on the site image is too hard for you.
@easyfruit5000 to be clear - in and of itself that doesn't cause me to leave a bad review either. Clearly specific types of prompts need to be followed to get good results with this model, and you need to follow a bit.ly link to learn how it's done. The bad review, then, comes from lack of instructions - and the lack of instructions come from the lack of following a bit.ly link. Was more my point, anyway.
@richpizor323 and again, the lack of the instructions is solved because the prompt is available on civit, just press the little circled i on each image
I tried version 2. I'm getting a lot of strange results. Model looks broken in version 2. Version 1 works much much better!
A V3 come today, I’m gonna add trigger word, fix strange bad body and more realistic !
Version 1 and 2 are missing their files
@malikxseto I removed them because some users will download them and give them a bad rating
@Shierry Did you the same with your old model Realistic RAW?
@malikxseto I have re-uploaded version 2 for you if you prefer. but it has a lot of bugs on V2 and there is no trigger word
@Shierry Thank you very much.. Even "failed" versions have their charm and can give interesting results in merging.
@malikxseto true like this new version is merged at 10% with v2



















