Trained on some hand picked scenes I personally liked. ( ͡° ͜ʖ ͡°)
From my testing v4 is very consistent and works fine at 1.0 strength, while v1 I had to set strength to 2.0 for consistent and better results, which sometimes distorted the image. Though v1 might still be a better choice when using multiple loras.
v4 info:
Samples were generated with lora strength at 1.0 and prompt:
a woman orgasming, she is experiencing an orgasm, a woman having an orgasm, she is orgasmingThe prompt is just a concatenation of the captions used for training this lora. It might be possible to get more specific results by mixing and matching them in a specific way.
v1 info:
Samples were generated with strength 2 and prompt:
A woman having an orgasmYou might have to test lora strength somewhere between 1 and 2.
Some example phrases that might help guide the model to better results:
She opens her mouth wide, she closes her eyes, her body twitcheshunyuan info:
I don't really use hunyuan but since a few people requested it I gave it a try. The results aren't nearly as good as the wan version.
I reused my wan dataset when training it, which is 16 fps, so it might look sped up at the 24 fps that hunyuan uses.
Description
FAQ
Comments (26)
Anyone else feel like she never quite gets to the end? It's all peak and no come down.
Yeah, I get what you mean. But it's kinda hard to fit it all into the 5 seconds that wan can generate reliably, since I also wanna include the buildup right before it starts, you know. ( ͡° ͜ʖ ͡°)
Maybe I'll try making a lora that would be trained for this. Then we could extend the video with it.
@abebes Please do, peaks and valleys!
Yeah, I do come down in a follow up video where she just has her eyes closed with a "serious face". It works pretty well.
@Clocksmith You using someone else's extension workflow or one you made yourself? I haven't gotten into successive generations yet but I need let her O finish....
@sarashinai I'm not using an extended video. When I consume my videos I have them on loops so having one smooth video isn't as important to me. So I follow up with the exact same seed so the look is the same, have the same sex position, etc. But when describing her face, I go from "orgasming" to "serious face".
@sarashinai I tried training a lora. I'm not quite satisfied with it yet, it is very inconsistent, so I won't be releasing it here.
Nevertheless, here's a hf repo if you want to try it out yourself: https://huggingface.co/uteresa/aftergasm/
Epochs 9, 10 and 11 seem to give usable results sometimes. Though I recommend adding "slow motion" as a negative prompt, as it tends to create very slow movement sometimes. For the prompt use "a woman calming down after an orgasm".
I also put there a simple workflow I used. It takes the last frame of a video to generate a new video and then appends it to the original one.
@sarashinai Also, this may not work with your own workflow but I also always have the "pingpong" option selected on my video combiner node, which creates a natural come down with the reverse of the come up.
@Clocksmith Will definitely try this!
FYI, this works for Wan T2V as well.
witch version ?
this is a waste of time don't try it works better without the lora
Hi, which version did you use?
When combining "hunyuan_v1" with sex and character loras, I find that a weight of about 0.6 or 0.7 is the maximum. Otherwise, the penetration / penis starts to disappear.
Please please please, make a lora for LTXV they released a trainer: https://github.com/Lightricks/LTX-Video-Trainer 😁
If you check out my generation with this LORA below (redhead with huge ta-tas) while I enjoyed the expressions generated, there was some strange interference and it messed up her face. Any way to avoid this? What did I do wrong?
Ever seen the movie Orgazmo? kek
For WAN t2v, I get better results with v4 at 2 strength than v1, even when using a character lora with m4crom4sti4.
Having great results with this!
This is such a fantastic LoRa. It works great by itself, but combined with others and properly prompted/weighted it really shines, adds an incredible amount of realistic motion. Really helps bring to life otherwise uninteresting generations. I use it all the time. Thanks.
The leg shakes are just.... chefs kiss
5/5, would recommend.
which version is for wan 2.2
I haven't made one, you can try playtime_ai's https://civitai.com/models/1814751/orgasm-wan-22-official-release
Can get all the body language just fine, but cannot for the life of me get any juice. Even tried kek's workflow and she's dry as a bone. Any prompt and/or weight help appreciated.
I usually combine it with the "WAN General NSFW mode" LoRA.
It usually does wonders for the squirting part, although it's best to hint in the image that it's coming etc.
@DrakPai Anything to be mindful of with the weights? I'd been playing around with combining NSFW, Orgasm v.4, and Cumshot together and was able to make progress but it sure doesn't take much weight for things to dry up again once Orgasm gets above say.. 0.3.