Wan 2.2 I2V LoRA for hardcore gay content — the image-to-video sibling of my hardcore-gay T2V LoRA.

Trained on the same 30-clip dataset, with the same trigger word and the same hyperparameters as the T2V original — only the base model and architecture differ. So if you already use hardcore-gay for text-to-video, this is the drop-in counterpart for image-to-video and first-frame continuation work.

Usage

• Trigger word: hardc0re-gay

• Base model: Wan 2.2 I2V-A14B (this will NOT work correctly in a T2V workflow — use the T2V version linked above for that)

• Strength: 1.0 is the default and what the examples use. 0.75 also works well if it is pulling the scene too hard.

• High/low pair: load high_noise into the high-noise expert and low_noise into the low-noise one. Crossing them over degrades motion badly.

• Camera vocabulary it responds to: closeup, far, sideview, topview

Notes

• Checkpoint 1250 of a 2000-step run. Later checkpoints over-train: from ~1500 the darker scenes lose detail, and by 2000 they artifact badly. If you train your own on this base, watch for the same turnover earlier than a T2V run would show it.

• Prompts work best in the terse, mechanical, image-describing style Wan prefers: subject → scene → motion → camera → lighting.

• 480x720 portrait at 81 frames is what it was validated at.

License

Use it, build on it, share merges — but credit me and don't sell it or what it makes. Merges inherit the same terms.