3 ms·
The selected examples got me thinking: Those eight are presumably examples the author felt turned out particularly well. Could one build a neural network traine
by mortenjorck 10y ago
The selected examples got me thinking: Those eight are presumably examples the author felt turned out particularly well. Could one build a neural network trained on generated ships that "turned out well" to automatically generate better-looking random ships?
Could this sort of process be used in games where procedural generation might otherwise be rejected because it looks "too random"?
- abritinthebay 10y ago> Could one build a neural network... Short answer: yes. Long answer: yes, but you'd have to retrain it for each style/form/concept that "well" was defined for. But yes, it's possible. Would be a lot of work though.
- KayEss 10y agoYou can think of the random possibilities as exploring one path through a huge tree of possibilities. You can then pick 9 paths for the next step and have the user choose the one they prefer and keep repeating until the user feels that they're done. We used to use this sort of approach to produce random shapes for animation and display on large screens in clubs in the mid-90s. You can also animate by tweening one or more parameters at particular branching points or between branches. If you pick the right places on these ships that could produce some interesting movement in the craft as well.