Comment by abetusk
4 years ago
Here's one that I think might actually have a chance of happening:
- Stable diffusion but for music. That is, give a text prompt and get a 2-5 minute song out.
So people can say things like "give me a song in the style of Siouxsie and the Banshee's first album, lamenting lost love" and get something reasonable back.
It would also be nice to do a retrospective of what actually was predicted correctly from past predictions, specifically 2022.
This exists, and uses stable diffusion: https://news.ycombinator.com/item?id=33999162
To me, this is not quite there. Stable diffusion produces, in some cases, really high quality output. The Riffusion is cool but still sounds a bit muddled.
Put it this way, if Riffusion came out in 2023, I would only give myself partial credit.
> - Stable diffusion but for music. That is, give a text prompt and get a 2-5 minute song out.
It already exists and it is called Dance Diffusion. [0]
[0] https://techcrunch.com/2022/10/07/ai-music-generator-dance-d...
[1] https://www.harmonai.org/
[2] https://colab.research.google.com/github/Harmonai-org/sample...
Already happening with Harmonai and riffusion. I think this year it may hit an inflection point with quality though. :)