Hey guys,

I'm sharing a "corruption" LoRA for MiniMax-H3.

I've always loved corruption / transformation / demon-girl themes, but they almost live in anime community. This time I brought it into a photoreal look. I have to say it turned out better than I expected.

This is a behaviour LoRA, not a "person" LoRA. Give it one prompt and anyone can go through the whole corruption arc in 10 seconds: black liquid parasitises her → horns grow → she awakens.

Of course, you can also get a "corrupted" look without the LoRA by just describing it in t2v. But trust me — apart from eating up a huge amount of time, it is not easy to pull off. I also put my own aesthetic into the transformation process, her expression, the black liquid, the horns and so on.

Well,this LoRA took me 100+ hours to prepare the dataset and train, I decided to share this with the community; if you like it, please show your support and leave a tip.

Hope u guys enjoy :).

---

Trigger word: corruptform

Base: MiniMax-H3 fl2va)

Resolution / length: 768×1024 (3:4 portrait) / 243 frames = 10.125 s

Sampling: 12–20 steps

Both pure text-to-video and first-frame conditioning work well.

---

## Prompt example

```

corruptform, Live-action cinematic realism, <camera statement, include "the framing never moves">;

she is <nude adult woman + pose + full body in frame>. <scene>.

At the start her body is completely clean bare human skin. Black liquid wells up out of the small glowing purple core on her chest.At the very end a pair of thick ridged glossy black horns pushes up through the skin behind

her hairline, and only then does she open her eyes again, her iris now bright purple.

```