Refusion vs Riffusion: Text-to-Audio Neural Composers

Refusion converts text prompts into high-quality audio spectrograms, blending neural diffusion and sound design.

Riffusion generates looping sound patterns directly from text using image diffusion trained on spectrograms.

Currently instrumental only, but integration with TTS systems is under development.

Yes, all sounds are uniquely generated, not sampled from existing works.

Experimental producers and ML artists exploring procedural audio generation.