I just tried to use Phenotype and noticed that it doesn't understand a lot of my inputs. Like 'shoegaze' or 'washed out' both gives you '?', only 'wet' makes it certain. the generated patch loads extremely fast tho, so i think it would be a fair trade-off to introduce a larger model that understands more language for slower loading times. Are you going to extend on that at some point? The basic idea of the feature is cool but I feel like if I can't talk to it emotionally, but have to spell out each feature of the patch myself (like modulated, wet etc), it kinda defeats the purpose a bit, because when i know what i want technically, i can also just go and make those changes myself

i discussed my findings about phenotype in this video:

and then i tried to use a larger LLM to produce a preset to test if that works: