MMH3 V1.0 (MiniMax-H3)

MiniMax-H3 can do slime girls decently well already with a good prompt and good starting image, but sometimes it takes a lot of tries, there's a position it's bad with, or the transparency just isn't quite right. This lora helps it be more consistent, but a lot of it still comes down to a good prompt. For some example prompts look through the showcase videos (the last 3 are T2VA). All of the showcase videos used either 20 or 22 steps with euler/simple, this lora at a weight of 1.0, the turbo lora at a weight of 0.6, and shift values of 6.0/4.0. Trained on FL2VA so it works for I2VA and gets like 80% of the way there for T2VA. I would recommend a helper lora or checkpoint that knows NSFW if you want to use this with T2VA. I haven't tested this with REF2VA yet but it should theoretically work at a lower weight. For T2VA or I2VA a weight of 1.0 seems to work well. Also I didn't train the audio but with enough steps and good enough prompting you can still get it to be accurate.

This is best used as a NSFW enhancer for slime girls rather than a standalone motion LoRA.

I got tired of waiting 10-15 minutes for a slime girl animation only to be disappointed by the transparency not getting animated right. In that regard this was originally meant to be similar to Translucent Penetration, but there is also some builtin motion enhancement.

The trigger word is slime_girls but you might not need it. A simple prompt like this works well enough:

He is enthusiastically fucking her ass. Detailed transparency distinctly shows his penis moving inside her body. She is very pleasured and expressive.

Or you can do something like this to try emphasizing better dynamic transparency:

Professional slimegirl animation. She passionately bounces on her ass on his penis. Detailed transparency shows her slime body moving around his erect penis. His penis can be seen inside her body tinted blue through refraction. His penis stays erect and rock hard as her gel body moves around it.

Don't describe the image for I2V unless it's getting important details wrong. Wan 2.2 works best when the prompt describes the actions in a straight and to the point manner.

For best results I would recommend using with SmoothMix Animations XXX (or something equivalent) and a checkpoint with good NSFW understanding. I usually use DaSiWa WAN 2.2 I2V 14B. The starting image also needs to have good transparency or it might not get the effect right.

Bonus tip - if you are using the lightspeed version of DaSiWa or another checkpoint that has the Lightx2v LoRAs merged, you can stack a high noise Lightx2v LoRA on top of your high noise LoRAs to boost the motion. Start with a weight of ~0.5. Higher weights give stronger boosts with higher risk of quality degradation and/or nonsensical movements.