C-LARA

I am getting quite close now to having the new image generation code integrating and thought it would be useful to have a toy text. It’s quicker to debug with; it also gives us some initial ideas about how we might do with picture dictionaries, which are unlikely to have the complexity you get when there are multiple related visual elements that need to be kept coherent.

Here is the automatically generated overview document for the classic movie plot Boy Meets Girl. I just entered the text (four pages):

The Age-Old Story: Boy meets Girl
Boy meets girl.
Boy loses girl.
Boy gets girl back again.

and gave it the style advice:

In a manga/anime style

The AI did everything else. As you can see, with two elements and a minimal plot, visual coherence is quite reasonable. Though, interesting, the AI is not at all satisfied with its work; when it evaluates the fit between the prompt and the generated images, it finds plenty of little glitches and assigns a low score.

More soon!