I leveraged OpenAI’s API as well as some other APIs to create an app that automatically creates videos and posts them to YouTube. I’ve been working on this project for a few weeks now, so I wanted to share what I’ve come up with. I initially set out to create children’s story videos, but because of how terrifying the generated images were, I decided to make it into a horror-theme instead. It seemed like the natural progression.
There’s still lots of improvements to make. Obviously, we’d love to get ‘themed’ images, but I’m not sure the DALL E API yet offers that functionality. Huge shoutout to @PaulBellow and @raymonddavey for helping me figure that out.
I’d love if you checked it out, and I would greatly appreciate any feedback. This project is still very much a work in progress.
Interesting idea, but the problem with AI generated art is the inconsistency between images. i.e. how your characters never look the same. In one image they’re young and in others they’re old. It would be good if it was “fixed”, perhaps the seeding could be retained at the user’s request.
thanks for checking it out! that’s the last major hurdle i would like to overcome. i’ve been trying to look into solutions, but i’m really hoping dalle eventually has some sort of image-to-image endpoint that we can use to create consistent images
Just a stupid question : how do you generate stories that long ?
With the completion limit of 4096 tokens, how do you tell Da-Vinci to provide a continued stories as several 4096 token completions ? Do you use a prompt like “continue the previous story” ? Did you used a fine-tuned model ?
Another question : what text to speech app do you currently use ?
i have no idea how they get so long. when i asked for children’s stories, they were much shorter. the length was a pleasant surprise when i switched to a horror-theme. my prompt starts with “generate a 1500 word horror story about…”