How can we provide reference images as anchors for facilitating reference subject based consistent image generation?
Codebase seems to compute anchor assets required (like DIFT features, last masks, etc.) in the first and second stage inference to provide as inputs for later stage inference on prompts other than anchor prompts.
How can we provide reference images as anchors for inputs and use these for generating rest of the batch images?
How can we provide reference images as anchors for facilitating reference subject based consistent image generation? Codebase seems to compute anchor assets required (like DIFT features, last masks, etc.) in the first and second stage inference to provide as inputs for later stage inference on prompts other than anchor prompts. How can we provide reference images as anchors for inputs and use these for generating rest of the batch images?