2025.11.24 added Wan2.2 I2V version

For the I2V version the key phrases are slightly different:

splits_pose, a nude woman is sitting on a chair, she starts doing the splits spreading her legs wide

splits_pose, a nude woman is sitting on a bed, she starts doing the splits spreading her legs wide, she lies on her back

splits_pose, a nude woman is lying on the floor, she starts doing the splits spreading her legs wide with her arms, lying on her back, high angle front view

Hi, this is my first lora, it’s for the woman’s flexibility fans – side splits.

At first, I tried to train a lora for Hunyuan video model (using diffusion pipe), but after spending a lot of time and money on a cloud service I didn’t get an acceptable result (maybe I did something wrong, you can give me an advice in comments).

Then I switched to Wan2.1 model and I’ve got an AMAZING result! (with the same dataset, process and settings as for Hunyuan one).

Trained on 31 videos 2-6 seconds length.

Trained resolution 832x464, 129 epochs.

For generation I’m using Wan14BT2VFusionX_fp16 model for better results with the following parameters:

Steps: 10-12

Sampler: euler simple, euler ancestral beta

Resolution I tried up to 1280x720 with good results.

Trigger word: splits_pose

Sample phrases:

splits_pose, European woman, brunette, a nude woman is sitting on a bed, she is wearing white stockings, she starts doing the splits spreading her legs wide, she lies on her back

splits_pose, European woman, blonde, a nude woman is lying on the floor, she starts doing the splits spreading her legs wide with her hands, lying on her back

Check out attached videos for detailed prompts.

P.S. As the “start” in the model name implies it’s the start phase of the movement and there will be other ones and not only splits, so stay tuned :)