Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by Vox_Leone | Hacker News Reader

Parent Full thread

Vox_Leone·

Can I use NLP to generate input for DALL-E 2? That would be cool.

View on HN

I used GPT-3 to 'write' a children's book and asked it to include descriptions of the illustrations.

https://docs.google.com/presentation/d/1y8EE_p8bw9dIEDguT1bT...

The fact that it's a derivative of an existing work is noteworthy, but I gave it absolutely no guidance on the topic. If i suggest something it will give it a go with similar fervor. eg https://imgur.com/a/N1qWaSV

Your link doesn't seem to be publicly accessible.

You can, in fact, use GPT-3 to engineer prompts for DALL-E 2 in a sense.

https://twitter.com/simonw/status/1555626060384911360

I want to see a few iterations of describing an image with AI, generating it, describing it again, generating it... Like when passing a piece of text through Google translate back and forth.

There was a tool that could find the "equilibrium" called Translation Party. I don't think it works anymore. I'd love to see one that goes back and forth between DALL-E and an image description algorithm.

I tried that! Results were mixed: https://twitter.com/pamelafox/status/1542593090472386561

It needs a better text to image model, I think. Maybe you can fork it and improve?

Interesting! I really like the flute > cup > bathtub sequence. It has a real dreamlike disjointedness to it.

According to internet popular belief, you'd end up with a picture of a certain ignominious dictator that unfortunately destroyed Europe in the 1940's. [1]

[1] https://en.wikipedia.org/wiki/Godwin%27s_law

Reply on news.ycombinator.com