Holly is made from prompting and Grok Imagine images, there were only 12 images in the dataset so she might feel a little stiff but i find she will follow prompts well at around 0.8-1.0