Trained on 2.2 I2V A14B model. Massively increased dataset over the 2.1 version.
Weight 0.7~1.0 ... 1.0 is usually fine
Some of the start images can be seen here: https://civitai.com/posts/21124019
Sample workflow in the 'training data' download. https://civitai.com/api/download/training-data/2129122
HIGH model is trained on the actual 'hip slammin' motion.
LOW model is trained only on ??????? close-ups.
My thinking is that this gives base Wan more freedom to work out her facial expressions in the low noise steps. The low noise lora only helps where it's really needed- the ????????... Whether this assumption is actually accurate 🤷♀️
a woman is straddling a man and eagerly ?????????? with him. she is squatting.(trained-in to the low noise lora)
The man's ????? slides in and out of view as it enters her ??????.His ????? is penetrating her ??????. you can see his ????? sliding in and out of her ??????.she moves quickly.she moves energetically.her ???? move quickly down on his ?????.she vigorously slams her ???? down on the man.she is quickly moving her ???? up and down. I also found that increasing the high noise CFG to ~6.0 helps a lot with movement on this lora.
Still not perfect- ????? often 'smooths out' losing any texture/veinyness. And sometimes it just totally breaks.
You can run just the HIGH version, but may end up with accordion ????? (????? getting smashed under her ???? rather than entering her ?????).
Since the lownoise lora is only helping with ????????, you can probably substitute other ????/penetration lownoise loras. No guarantees.
1. Dieses Modell dient nur Lernzwecken. Urheber- und Auslegungsrechte liegen beim Originalautor.
2. Bist du der Originalautor eines Modells, kontaktiere uns bitte zur Authentifizierung über unsere offiziellen Kanäle. Wir schützen die Rechte aller Schöpfer. Hier klicken, um es zu verifizieren.
