After the huge demand following my previous Futa Cock Growth LORA a lot of you asked for a simple, no-frills realistic penis that actually looks good on both women and men. This is my first public attempt. It’s surprisingly decent for a v1, but still limited in angle/size/shape variety. I’ll be retraining soon with a much bigger dataset for v2.
Works on:
Females (futa/tgirls)
Males
Solo or couple scenes
Sometimes can do different skin tones but its hit and miss
Base model: Z-Image Turbo
All sample images: Z-Image Turbo, Euler, 9 steps, CFG 1.0, strength 0.4–1
v1.5 is better at lower strenghts vs v1. Also works really well along side Mystic XXX for female anatomy fixes.
☕ 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
FAQ
Comments (15)
I have a big realistic futanari dataset if you want.
Hi, please DM me the dataset.
Great work! I'm getting great results with smaller weights '0.3/0.5'. In the next update, could you add more images in the 'POV' view? I'm not having much success with only this view, and it's the view we use most often. Thanks!
That's interesting, I didn't try any weights below 0.8 in testing. Thank you for the feedback, I'll add some more POV images to the dataset for the next run. How are you finding size control or do you not care most turn out massive haha
@BalakayMotion Great! I'm using that 0.3/0.4 power because I'm using it in conjunction with the amateur realism Lora (https://civitai.com/models/1662740/lenovo-ultrareal) which gives me incredible realism results, and with these values it doesn't interfere with the results even if your Lora is a bit '3D/render style'. Since Z-Image produces penises without Lora even if deformed, then your Lora finishes giving the necessary fine details with just this minimum power. As for the size, everything is ok, haha ^^. Looking forward to seeing more updates of your great work, thanks!
Noticed that on strength 1.4 it completely breaks hands, most only have three fingers and that with the sample prompt for the girl with penis...
which version of the model are you using?
@BalakayMotion v1.0, only version here atm
@Kaleidia Sorry, I meant which Z-image model.
@BalakayMotion tried with the normal turbo here by the ally, fp8 and q8, same result
Realistic models like z-image works better with light strengthened LoRa. I have decent results at 0.3 with this one. All my LoRa are between 0.2 and 0.3. If you don't have decent results maybe you should try to improve your prompt. ZiT is very powerful with natural texts rather than keywords. Good luck !
This LoRA appears to influence facial features, particularly when generating Asian males or overweight males, resulting in noticeable differences compared to outputs without it enabled.
try at a lower weight, some have said low works well like 0.4 (not tried myself yet) its odd that its changing any facial data as non of the dataset contains a face
Thank you! Can I ask how many images, steps and which Learning Rate you used for the training? Thanks!
Working Great. In the next update could you also add Anime, Cartoon, 3D style Loras. flaccid and and soft variants too. Please.