Rendering Text on Poster Images with the Original Image

[k00dj-19]

Currently, the framework is designed for text-based poster image generation.
However, I am wondering if there is a way to extend this framework so that when an image is provided as input, the framework only renders the text in an aesthetically pleasing style without altering or changing the original image content itself.

[Ephemeral182]

Thanks for your attention. You can check out some recent methods for reference control based on MM-Dit, or try fine-tuning to build an edit-based model.