This is the Version 1 release of my Cumshot LoRA for MiniMax H3, designed to add a visible cumshot sequence to your video generations.

For the showcase, I used still images of erect penises in a variety of scenes. I used Ollama with Qwen-VL to generate prompts for 10-second videos, asking it to make something funny happen while the camera follows the path of the cumshot. The prompts followed the original creator’s recommended guidelines.

The showcase was created using my quick day-one prompting setup. Custom prompt-generation nodes are now available and will probably produce more consistent results. Workflows are embedded in most of the showcase videos, allowing you to inspect the settings and reproduce the examples.

Version 1

Version 1 was trained for several additional hours, reaching 7,000 steps. I’m currently updating the showcase, although the results have not changed dramatically from the earlier 5,500-step preview release.

Depending on your source image and intended result, you may want to pair this with my Penis LoRA at a low strength.

Troubleshooting and Feedback

If you encounter problems, please begin by trying to recreate one of my showcase examples using its embedded workflow. Whether that workflow works on your system will tell us a lot and make troubleshooting easier.

If you can reproduce the showcase but have trouble with your own images or prompts, let me know what happened and include your settings if possible.

I’m always open to suggestions for improving my models. If you own suitable sample material that you would like me to consider adding to the training dataset, please reach out. Concrete examples are much more helpful than general suggestions.