search

ai-forever / Kandinsky-2

Kandinsky 2 — multilingual text2image latent diffusion model
Apache License 2.0
2.75k stars 307 forks source link

How to add text to an image #81

Open Sankalp96 opened 1 year ago

Sankalp96 commented 1 year ago

I'm unable to add text to the image. Is it possible to add text and it automatically figures out the orientation and layout of the text and fits in accordingly in image?

adamfils2 commented 1 year ago

Text-to-image diffusion models are bad at generating text in images because they are not trained on a dataset of images that contain text. These models are typically trained on a dataset of images that do not contain any text, so they do not have the ability to learn how to generate text in images.

In addition, text-to-image diffusion models are typically trained to generate images that are realistic and visually appealing. This means that they are focused on generating images that have the correct colors, shapes, and textures. They are not focused on generating images that contain text, so they may not be able to generate images that have text that is readable or understandable.