TwinkCockTNG
This is my ninth major publicly posted LoRA. This is my second Klein 9b LoRA and using AI-Toolkit.
Please click on "Show More" under "About This Version" to get strength, sampler and scheduler recommendations.
This is essentially a complete rebuild of the TwinkCock series, incorporating newer "lessons learned" and newer captioning strategies from my recent LoRAs.
The current version is an alpha. It has idiosyncratic strengths and weaknesses.
A wide range of images were used for this LoRA. A total of 25,590 images and 2,616 regularization images were used in this data set.
The LoRA is strong in generating an erect penis of average to above average size, and was also trained on masturbation and ejaculation concepts. It generates penises more consistently in the reclining, sitting, and kneeling position. It struggles with generating penises in a standing position, and penises that are flaccid, semi-erect. It is not responsive to changes in penis size. It confuses circumcised and uncircumcised penises.
Like my other models, it is okay. Please continue reading for setting recommendations:
Summary for LoRA:
This is a concept LoRA that will generate a penis, most often in an erect state. The goals of this LoRA were to:
Recreate the TwinkCock dataset series for Klein 9b and prepare to train newer generation LoRAs.
Extend the LoRA to include masturbation and at least two men engaging in masturbation.
Extend the LoRA to include concepts from the FlyingCum series.
Test newer captioning strategies including Gemma 4, Qwen 3, and JoyCaption Beta.
Tags were manually edited but to a lesser extent than previous LoRAs.
The primary activation tag "twinkcockTNG" was added to all images. A number of secondary tags (see below) were added depending on the subconcept but these appear to be less needed than Flux1.
Other notes:
For images depicting ejaculation, explicit description of the target of the semen likely needs to be included, Klein 9b's default appears to be the ejaculator's mouth.
The semen appears to be flexible and modifiable. Use with care, and the user assumes all responsibility in complying with all guidelines for posting.
Black African and East Asian examples were included in this data set. Klein 9b's default for Asian generations appears to favor a tanned complexion.
Age representation will likely rely on the main Klein 9b model. Most images used for training were twinks.
During training, the captioner was asked to classify the man in terms of "ectomorph", "endomorph" and "mesomorph" bodytypes.
Women were not included in the training data, and I have no clear idea what will happen if a women is specified.
Regularization images were used. Extensive testing on style flexibility beyond photorealistic has not yet been conducted but compatibility with style LoRAs appears successful.
Secondary tag list (optional, mostly here for reference):
CockThruFly: A cock through an open fly. (Describing the fly opening is often helpful).
CockThruLeghole: A dickslip through the leghole.
BetweenPantsAndShirt: A dickslip between a shirt and pants.
NoPants: self-explanatory.
NoShirt: self-explanatory.
FullyNude: self-explanatory.
AsianCock: East asian examples.
1ppl: Scenes depicting one man.
2ppl: Scenes depicting two men.
3ppl: Scenes depicting three men.
CumReclining: A cumshot laying down.
CumSitting: A cumshot sitting down.
CumStanding: A cumshot standing.
CumSitting: A cumshot sitting.
CumHandsFree: A cumshot explictly handsfree.
CumSelfer: A cumshot explicitly aimed towards the mouth of the ejaculator.
Special thanks to @markury and the members of the Bulge Discord server https://thebulge.xyz for their support, advice, and beta testing.
Description
LoRA Strength: 0.8-1
Sampling Steps 11-13
CFG scale 2.25
Sampler: DPM++ 2s a RF, Euler a, LCM
Scheduler: Flux2, Bong Tangent
Comments (2)
your lora's are an instant download for me. Such a welcome addition! thanks for your work you put into this. Appreciate it.
25,590? Never heard of that many. I cut mine down from 600 to 200 and seems fine. It is because you have so many concepts in there?















