Wan 2.2 I2V Creampie LoRA - Cr34mp13
My first attempt at training a Wan 2.2 Image-to-Video (I2V) LoRA, showing promising results for NSFW workflows. I haven't had much change to test it properly so post away!
Note: Trained High Noise Lora to 80 Epochs and seems to have gotten rid of the liquid coming out of the mouth issue, currently training Low Noise and will update it when it is done.
Overview
This LoRA is designed to be used in the second part of a full NSFW workflow for merging two videos - one video starts the NSFW act, and this LoRA helps with the finishing/creampie effect. The main objective of the Lora was fora strong Creampie effect during penetration. Seems to work for Anal and Oral Creampie as well. Further testing is required.
Strength: 0.6 - 1.0
Training Details
- Dataset: 79 clips (2-8 seconds each)
- Resolution: 320 x 192
- Content Mix: 70% realistic + 30% anime
- Anime addition helps with exaggerated action and fluid amount
Prompts
Vaginal Creampie
cr34mp13, Thick white cum is pouring out of her vagina and around the penis. Lots of thick transparent white cum flows out as the penis goes in and out of her vagina.
Anal Creampie
cr34mp13, Thick white cum is pouring out of her asshole and ass around the penis. Lots of thick transparent white cum flows out as the penis goes in and out of her ass.
Notes
- All examples done with LightxV2 Speedups
Support
If you like what I'm producing and would like to support me and help me make more awesome Wan 2.2 I2V Lora's, a coffee can go a long way :). Also will do custom requests, just hit me up.
Thanks!
Description
FAQ
Comments (8)
It would be great if you posted your yaml file for training
Hey of course ill do that when I get home
Uploaded the Training Data
@aivantgarde thanks - I tried using the community proven settings but I am having no luck. My loss just isn't going down - I plateau immediately and it's like that for 1000 steps (high noise).
I could not get this to work in 10 tries. It either does nothing or for 2 of the gens it made her vomit.
It works for me. I've only used it with Lightxv2 loras for speedup. I've found using the v1.0 low noise with the 1.1 High noise produces more exaggerated effects. Ill do more isolation testing when I get the chance.
Setting the strength to 0.6 seems to work more consistently.
@Fug0 Ill reuploaf the High Noise V1 as I think it provides more exaggeration seems training further it has adapted more to the real samples
Details
Files
CRM-FULL-EPOCH-80-LOW.safetensors
Mirrors
CRM-FULL-EPOCH-80-LOW.safetensors
CRM-FULL-EPOCH-80-LOW.safetensors
CRM-FULL-EPOCH-80-LOW.safetensors
CRM-FULL-EPOCH-80-LOW.safetensors
CRM-FULL-EPOCH-80-LOW.safetensors
WAN-2.2-I2V_Action_CreampieCRM-FULL-EPOCH-80-LOW.safetensors
CRM-FULL-EPOCH-80-LOW.safetensors
CRM-FULL-EPOCH-80-LOW (1).safetensors
Creampie CRM-FULL-EPOCH-80-LOW.safetensors
CRM-FULL-EPOCH-80-LOW.safetensors
NOTUPLOADED-CRM-FULL-EPOCH-80-LOW.safetensors
CRM-FULL-EPOCH-80-LOW.safetensors
Wan2.2-I2V-Creampie-INSIDE-LOW.safetensors
CRM-FULL-EPOCH-80-LOW.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.