This embedding is trained with 41 spectrogram images with descriptions from my song : https://soundcloud.com/wedgeewoo/sore-wa-wonkyfuntime-da?si=81954182cc7446fe8be468def7f03277&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing
Here's an example of what it can produce. https://soundcloud.com/wedgeewoo/riffusion-test-minimal-tech-house-with-sub-bass-reese-bass-by-sorewaplusv2?si=97612027601b4ea48ab95a51a688ff2f&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing
Prompt : minimal tech house with sub bass, reese bass
The same seed, but a random seed for the variation option with some strength.
Also trained with descriptions of 150 subgenres of EDM.
v1 was over trained, stepping back to 15,000 steps.
Prompt example : neurofunk EDM song with drums, guitar , piano. reese bass, by wedgeewoov2
Description
This is the latest version of the wedgeewoo embedding, stay tuned for more in the future!
In this embedding i used the embedding editor extension in the Automatic1111 and adjusted some of the weights to get some good results.
I think i just got lucky adjusting the weights, no rhyme or reason it just works