Show HN: Turn text into images with VQGAN+CLIP

justswim 2021-08-18 01:10:10 +0000 UTC [ - ]

Hi Hacker News!

There was a recent post on Show (https://news.ycombinator.com/item?id=28120247) for a collab notebook that allowed anyone to create AI art. We took that code and made a hosted version that makes it easy to get a video with just a query - no setup or coding experience necessary. Just click to the page and you’ll be able to create new art for free.

The generator is based off of the notebooks (https://colab.research.google.com/drive/1n_xrgKDlGQcCF6O-eL3...) written by Hillel Wayne (https://www.hillelwayne.com/) and Katherine Crowson.

You just type a query and you’ll get both an image and a video that the gan will generate. Because it’s seeded randomly each time, no two videos will ever be the same even with the same query.

Would love to get your feedback or see what you make!

minimaxir 2021-08-18 01:18:45 +0000 UTC [ - ]

Calling it an "AI Video Generator" is IMO misleading; it's a video of the intermediate frames during the training process as it improves, the AI itself doesn't generate the video.

justswim 2021-08-18 01:37:19 +0000 UTC [ - ]

That's good point ... I just figured it was a little bit more of an inclusive term since it does handle stitching the frames together. But agreed that it could be more clear.

motyar 2021-08-18 12:46:55 +0000 UTC [ - ]

Oh that's different, I thought it converts text into images like https://bruzu.com does.