<h2 id="concept">Concept</h2><p>A woman wearing a strap on dildo penetrates a man anally. Minimax actually makes a shockingly good go of this even without the lora, but it often gives you giant death dildos or the motion is wrong. This lora can generate male genitalia but it's not the core focus, and often you will get imperfections. All the examples only use this lora.</p><p></p><p>If you want perfect dicks i suggest you add a cock lora or XXX lora with too, or start with the man visible in the first frame. </p><h2 id="usage">Usage</h2><p>Should work in both T2V and I2V. I would recommend a weight of 0.5-0.8. Minimax requires detailed prompts, but here are a couple of good starter prompts that work well out of the box. I would highly recommend you look at the prompts for the samples as they should work well with most images.</p><p></p><pre><code>The woman puts her hands on the man's hips and inserts the dildo into his anus. As she's inserting the dildo she feels resistance and has to adjust to insert it fully. Once inserted she thrusts her hips back and forth pegging the man, as she thrusts the man's penis and scrotum swing slightly from the impact. 

overall_soundscape: The noise of skin slapping skin as the woman's thighs make contact with the man's buttocks. The grunting of the man as he is anally penetrated and the hard breathing that follows.</code></pre><p></p><p>The one major concept it doesn't do is male POV, which is largely because i used the same original dataset as the WAN lora. I'll probably find some sample clips and re-train this concept into the lora in the future.</p><p></p><h2 id="training">Training</h2><p>The model was trained on 65 videos, a mixture of 3rd person and female pov. all 5 seconds long and 24 fps with audio. It's trained at about 0.6mp and 6 frame buckets for about 2900 steps. Captioning was automated using Qwen3 VL 32B heretic. Training was run using diffusion pipe on 3x3090s.</p><p></p>

Files with SHA256: 67c5f697b9a80a01af925a29929450afd3d42b42a7908622218d7e14a14d6057