Upload 3007557/3412628/README.md with huggingface_hub

3007557/3412628/README.md

@@ -0,0 +1,63 @@
+---
+license: other
+tags:
+- action
+---
+
+# Minimax Softsuck Blowjob - v1.0 - 3412628
+
+**Model Type**: LORA
+
+**Base Model**: MiniMax H3
+
+**Trigger Words**: None
+
+**Tags**: action
+
+## Gallery
+
+<table>
+  <tr>
+    <td><video src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e8459028-409b-47e9-87c6-8ac07c5c8a7e/original=true/146003077.mp4" width="200" controls muted autoplay loop></video></td>
+    <td><video src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/986e11aa-1589-46c3-8196-75c39b85fdc5/original=true/146003097.mp4" width="200" controls muted autoplay loop></video></td>
+  </tr>
+</table>
+
+## Description
+
+Concept
+-------
+
+A minimax version of my WAN soft suck lora. Minimax doesn't do great with flaccid cocks leave alone blowjobs on flaccid cocks. It can generate cocks but i will say its far from perfect and can end up with body horrors/detached cock syndrome. This is an experimental lora from my point of view, see the training notes for more details on why.
+
+Usage
+-----
+
+Should work in both T2V and I2V. I would recommend a weight of 0.7-0.8. Here are a couple of good starter prompts that work well out of the box. The model was trained on cock size but doesn't always do great with that.
+
+It works fine with both POV and side angles. Take a look at the prompts in my samples for examples of how to get that working if you aren't starting with a cock in the scene.
+
+### Prompts
+
+```
+She takes the glans into her mouth and begins soft, rhythmic suction, her head moving in small controlled forward-and-back motions
+```
+
+I would highly recommend you start with the prompts in the samples instead of just this.
+
+Cock keywords it's been trained on are flaccid, large/small for size and uncut/circumcised for state. If you're introducing a cock into the scene use those. It's also been trained on POV shot or man's POV so will adjust to that if you add those words.
+
+Training
+--------
+
+The model was trained on 73 videos, 38 POV and 34 side view, all 5 seconds long. The reason this is a experimental lora was i decided to be lazy and see if could just reuse my dataset from my WAN Lora which is at 16 fps. So I'm still not sure if that's had an impact. I can't see anything visually/obviously wrong in the gens but maybe I've just not pushed it enough. Very keen for feedback if people find that training with 16 fps input videos had a negative effect on the quality/animation. It's trained at about 0.6mp and 3 frame buckets for about 6000 steps. Captioning was automated using Qwen3 VL 32B heretic. Training was run using diffusion pipe on 3x3090s.
+
+---
+
+Author: [imb101](https://civitai.com/user/imb101)
+
+Model: [CivitAI Model Page](https://civitai.com/models/3007557?modelVersionId=3412628)
+
+Archive: [CivArchive Page](https://civarchive.com/models/3007557?modelVersionId=3412628)
+
+<!-- Version: 20261018_upload -->