This gallery showcases a collection of characters generated from text prompt with consistent poses (the poses are different for females and males). If you click on any character, you can see front and back view. The goal of this project was to help concept artists brainstorm the ideas for the outfits quickly within Ready Player Me company.
The project uses customly trained Stable Diffusion v1.5 model. Originally, the character style was derived from Ready Player Me characters' renders mixed with some other style that helped to enhance the quality and diversity significantly.
More details on the implementation you can find in the
blogpost. I also encourage you to generate some characters using Replicate API. Some keywords for most of the characters are available if you click on them.
Blogpost |
Replicate API |
Github