Trained using AI-Toolkit on 20 hand-trimmed clips, 73 frames at 24fps, with their real audio. 1,600 steps, rank 16, 512px.
Trained on , Cowgirl, doggy.

Works great for any scene with I2V.

I DID NOT test T2V.

hmmotion, [position + who is doing the work] + [hard and fast / slow and deep] + [what visibly moves]. [the wet sounds] + [her voice].

Example:

hmmotion, She's  hard and fast, bouncing up and down on his , her  jiggling with every thrust. She's gasping and  loudly, a high-pitched "ah, ah, ah" with each bounce, over the wet, sloppy sounds of her  on his .

Blunt words work, clinical words do not. Use , , , , -deep, wet, sloppy, squelch. Skip anything that sounds like a medical report.

Trained at 512px as a proof of concept, so treat it as experimental. A higher-resolution version is next.

You can deploy MiniMax on Runpod using my Runpod template:
https://get.runpod.io/minimax-template