sd-concepts-library/stable-diffusion-conceptualizer · Tips and examples on using a person's likeness in generation (For putting yourself in a Halloween costume!)

A concept <selezneva-alisa> contains a likeness of a character unknown to 1.4 SD model. For people who want to use someone's likeness in generation, this is a good test case. There are 2 ways:

Find a prompt (e.g. on lexica.art) that leads to an image of a person with correct anatomy, and replace the character's name with <selezneva-alisa>

Prompt: VFX movie still frame portrait beautiful DC vs. Marvel hero <selezneva-alisa> natural skin, hero pose, natural evening light in the city by Emmanuel Lubezki

Inpainting. There doesn't seem to be one on Hugging Face. Install SD on your own computer e.g. from https://github.com/AUTOMATIC1111/stable-diffusion-webui , choose an image you want to replicate, then use inpainting on the face with <selezneva-alisa> in the prompt:

In both cases the character is barely recognizable, any suggestions on how to improve the faithfulness to the original training set are welcome!