This LoRA responds well to "penis from behind, ass toward viewer" prompts and is my go to for this style but it's also a regular Penis LoRA that isn't just rear view so go wild and find some front view, side view results too.
Trained with more images than bigpp_v2_SD1.5 also using a Sony a6400 Camera. Noticed more veins, fuller and more detailed testicles and the penis head looks excellent since I took pictures at over 100 different angles and views.
Step 1
Trigger Words: bigpp, big penis, erect (flaccid works also)
For Rear View Prompts: mine will look like: "bigpp, (big veiny penis from behind, flaccid, perineum, testicles from behind), ass towards viewer"
Weight: 0.5-1
0.6-0.7 for best results - ***in step 2, you can then go 0.8, 0.9 or even 1, to generate an insanely detailed penis***
Model: yogirimix_v10 (used most, best penis definition), DarkSun works too. Might work with MatureMaleMix, FefaHentai, RealCartoon3D_v16, DarkRevPikas, and I've used it with many others others.
Clip Skip: 1 or 2 (use 2 if you're not getting a penis but rather just ballsacks or a deformed vagina)
VAE: vaeFtMse840000EmaPruned
Sampling Method: DPM++ 2S a Karras (more detail in penis, especially the head, veins, or ejaculate), DPM++ 2M SDE Karras (faster, sometimes better penis positions or views), AND then Eular a Karras sometimes surprises me with some of the best penis outputs I've ever generated.
Sampling Steps: 25-65 depending on CFG
CFG: 6 with 25 steps, 7 with 35-40 steps, 8 with 45 steps, 9 with 50-55 steps, 10 with 65 steps
Resolution: 512x768
Step 2
Send to: img2img
Sampling Method: (all using Karras):
a). DPM++ 3M SDE (used most)
b). DMP++ 2M SDE (for results that doesn't change the look of the penis too much)
c). DPM++ 2S a (for more variation in tones, lighting, shading, etc.)
d). Eular a Karras if you did Eular a in step 1 but you can still use other samplers on a Eular a img.
Sampling Steps:
Same as last step, or you can adjust higher. Say you did CFG 7 with 35 sampling steps in the last step, you can go CFG 9 with 50 sampling steps or whatever you want here to get greater detail. I like 65 steps with CFG 10, 75 steps CFG 11 or 100 steps CFG 10-12.5
DON'T FORGET TO RESIZE: from 512x768 to 1024x1536
Prompt: in step 1 I like to use a LoRA weight of 0.6-0.7, but here in step 2 you can do a weight of 0.8-1 for more detail in the penis but will look closer to the Realism of the training images. I like 0.8-0.9.
Denoise:
a). 0.17-0.2 to keep results close to exact as original image
b). 0.25 for close to same as original but if you don't like some detail on the penis, this will generate a small fix
c). 0.3 for still same as original with minor detail changes
d). 0.4-0.45 for some changes to original and better detail
e). 0.5 for randomized changes, insane detail
f). 0.55-0.65 more random change but more anomalies, but even crazier detail
g). 0.69 anomalies might balance out, resulting in a greater change or fixes in ie. "deformed limbs" in the original, and even greater detail as shown in this example:
Then you can generate and the resulting image will be pretty good, but you can make it better by doing this:
Step 3
Send to: Extras, select the 4xUltraSharp upscaler, resize by 2, DONE!
Sometimes the img2img (step 2) looks better (better detail in hair, pubic hair, eyes) than step 3 but more than not the 4xUltraSharp generates the best quality final result. You can download 4xUltraSharp here: https://civarchive.com/models/116225/4x-ultrasharp