Concept

A good old JOI hand gesture lora, kinda shocked one doesn't exist already for minimax.

Usage

Should work in both T2V and I2V. I would recommend a weight of 0.6-0.8. At higher strengths you might find the audio starts to get very whispery, given much of the source content involved women talking in hushed tones. I tend to find at higher resolutions, in particular 2mp, a higher weight (0.8) helps.

Minimax requires detailed prompts, but here are a couple of good starter prompts that work well out of the box. You can chain them together to get the effect you want, or let the model figure it out with just the base prompt.

Base prompt

You can change the hand obviously. The hand does not need to be in the original frame for it to work / appear.

she begins to gesture with her right hand, in a stroking motion, mimicking masturbation. 

Hand shape

Try different things here, its largely driven by the base model which can do a lot

She makes a loose fist with her hand, the thumb and forefinger connecting in a circle. 

She makes a loose fist with her thumb and forefinger making a wide circular shape. 

Hand movement

You can pick a single location for her to focus on

Her hand moves in a rhythmic, mimicking motion near her crotch, simulating male masturbation with a steady, deliberate pace.

Or you can use Minimax's knowledge of human anatomy to determine where she moves her hand from and to. Just replace the from/to parts in the below prompt and add it in.

Her hand moves in a rhythmic, mimicking motion. Her hand starts near her navel, moves downward to her crotch, then returns upward in a continuous motion. 

Speech

Obviously its no JOI without the instruction. Be aware that the tone and text you provide can make differences to how she moves as well.

As she moves her hand she counts down and says in a commanding tone <d>[English] Five.. Four.. Three.. Two.. One.. Come for me.. </d>

As she moves her hand she says in a commanding tone <d>[English] Jerk faster bitch.. i want you to empty those balls.. </d>

As she moves her hand she says in a subservient tone <d>[English] Please master.. bless me with your seed </d>

Training

The model was trained on 40 videos. all 5 seconds long and 24 fps with audio. It's trained at about 0.6mp and 3 frame buckets for about 2720 steps. Captioning was automated using Qwen3 VL 32B heretic. Training was run using diffusion pipe on 3x3090s.