Typical prompt
bl0wb4ng, one woman and {two|three|four|five|six} men, {blowjob|handjob|switching}, {static|dynamic} shot, ...
Example
Unused words can be added in the Negative prompt to increase adherence.
Positive: bl0wb4ng, one woman and two men, handjob, ...
Negative: three men, four men, five men, six men, blowjob, switching, ...
Usage Tips
If you don't know where to start, the first showcased video contains an easily reproducible workflow, with as few custom nodes as possible. Just download the video from its dedicated page, then drag-and-drop it into the ComfyUI canvas to display the workflow.
The dataset is distributed around four men, with two and six being the least represented number of people. Wan will usually draw them with a precision of +/- 1 person, it's not 100% reliable.
It can draw a progressive transition instead of a cut, by setting the strength of the high-noise lora to ~0.5-0.65 and adapting the prompt accordingly (setting the number of men to two seems to work…
Tweaked a few knobs and pushed the training a bit further.
🟢 Less hallucinations
🟢 Better hand skills
🔴 Slightly reduced facial accuracy
Usage
Prompting: keep it simple; avoid NSFW vocabulary when possible.
Shift: keep it low. 7 is OK with most of schedulers and evenly split steps.
Cut: the cut happens at the 9th frame.
Lower Strengths (high-noise):
~0.5: makes a progressive transition instead of a cut.
~0.7: less crowded scene.
Misc
All the showcase videos have their workflow embedded.
Tested only:
with lightx2v LoRAs enabled (1030 or 1022 on high-noise and 1022 on low-noise).
in ComfyUI.
on the following base wan models: Q4, Q5, Q6 and nvfp4.
Training
v1.0: 67.5 epochs, 72 5-seconds clips (4860 steps).
Blowbang concept LoRA.
Trained on 21 clips of 5 seconds, 16fps.
Use this
A woman is surrounded by multiple penises, she is in the middle of the frame performing oral sex, men's torsos are around her, they all have huge penises which the woman are grabbing and putting inside her mouth.
Then customize to your liking.
NOTE: Not the result I was hoping for, but it is a very hard LoRA to train, especially with all the movement the woman is making by shifting the way her head is moving from side to side. Maybe an improved dataset where the woman is focused on one penis while the other penises are just surrounding her would yield a better result, for next time hopefully.
Workflow is included in the samples.
If you enjoy these models and find them useful, please leave a thumbs up on both. It really motivates me to keep creating more! Feel free to also share some of the work you generated with them in this page; I’d love to see it.
