Custom-trained Wan 2.2 I2V LoRA on curated female height growth clips. Distilled from other forms of growth. May not work on male subjects.

The model is trained on natural language size increases using numeric keywords between "1.1x" to "1.9x", e.g. "A woman grows 1.4x taller", "An anime girl grows 1.7x bigger", "A person grows to 1.2x their initial size". Trained on both anime and realistic styles.

The result is usually not accurate to the prompted size. It instead mainly enables Wan 2.2 I2V to output a variety of target heights. Here are some things you can do to influence the output height:

• Rule of thumb: Faster noise falloff = More growth

• Using Euler sampler, the following schedulers produce from least to most growth:
  Linear Quadratic | Normal | Beta.

• Higher sigma shift = More growth

• Higher weight for LOW LoRA results in more contrast between 1.1x and 1.9x.