AI AI agAIn

[barnoid]

8 years ago, for NaNoGenMo 2016, I used image captioning AI on 5036 stills from the movie A.I. Artificial Intelligence (2001) to attempt to generate a movie novelisation.

Now, for better or worse, I'm going to try this again using current AI tools to see how things have changed.

I suspect the text will be considerably more coherent and more clearly based on the images, but I doubt it will be any easier to get an idea of the film.

[barnoid]

Progress:

• I ripped the DVD again, using ffmpeg to extract stills directly from VOBs seems to fail.
• Used ffmpeg to extract 1054 images from the video.
• This time I'm using the chapters on the DVD as the book chapters.
• I'm using Llava 1.5 from here https://github.com/Mozilla-Ocho/llamafile running locally.
• For the first image in a chapter I give a prompt and the image to Llava and record the resulting text.
• For the rest of the images I give it the prompt, the previous image's text and the image.

So far 20 of 32 chapters are done for 40,824 words. I'm reading it all as we go to make sure it doesn't say anything horrible. If I don't like the chapter text I regenerate it.