RAVE Latent Morpher
Upload two sounds. The model encodes both into latent space and generates
a smooth acoustic path between them — each step is a mathematically interpolated
point, not a crossfade.
Set Model A ≠ Model B to enable cross-model morphing: Sound A is encoded
in one model's space, Sound B in another's. The decoder navigates between
two completely different learned representations of sound.
Best results: solo instruments or voice, 10–30 sec.
Mixed/produced music sounds muddy — these models were trained on single timbres.
First run takes ~2 min to download models.