Riffusion

The spectrogram experiment that turned into a song platform.

In short

Riffusion began as an experiment generating music by diffusing spectrogram images and has since become a full song generation platform with vocals. It retains a distinctive sonic character and is available directly and through model aggregators.

Riffusion specifications

Output: Music with vocals
Origin: Spectrogram diffusion
Access: API, Replicate
Character: Textural, experimental
Status: Evaluating

Figures verified 2026-08-10. This field moves quickly — re-check before relying on them.

About Riffusion

Riffusion has the most interesting origin story in the field and it still affects how the output sounds. The original project fine-tuned an image diffusion model on spectrograms — pictures of sound — and converted the generated images back into audio, which was a genuinely clever hack rather than a purpose-built architecture. The current platform is a far more capable song generator, but a distinctive character has survived the transition, and a segment of users prefers it precisely because output does not sound like everything else. The lineage is unusually well documented, which makes it valuable for anyone trying to understand how these systems work rather than just use them. Availability through aggregators alongside direct access means it is easy to try without another subscription, and that low friction is a large part of why it stayed relevant while better-funded competitors disappeared.

Strengths

What Riffusion is good at

Distinctive character

Output has a recognisable character that does not sound like the homogenised middle of the category, which some users actively want.

Well-documented lineage

The spectrogram-diffusion origin is documented publicly in unusual detail, making it genuinely useful for understanding how these systems work.

Available on aggregators

Access through aggregators means you can try it inside a workflow you already have rather than starting another subscription.

Riffusion limitations

Less polished than the leaders on mainstream genres
Smaller ecosystem
Closed current weights

How it compares

Choosing Riffusion

What comes back

Riffusion began as an experiment generating music by diffusing spectrogram images and has since become a full song generation platform with vocals. It retains a distinctive sonic character and is available directly and through model aggregators.

Compare with MusicGenerate

Riffusion review, pricing and alternatives

Searched as riffusion, riffusion ai, riffusion ai music and riffusion classic. The spectrogram origin explains why its output has a character people either seek out or avoid.

Riffusion — questions people ask

How did Riffusion originally work?

It fine-tuned an image diffusion model on spectrograms — pictures of sound — and converted the generated images back into audio. A genuinely clever hack rather than a purpose-built architecture.

Does Riffusion still use spectrogram diffusion?

The current platform is a far more capable song generator than the original experiment, but a distinctive character has survived the transition, which is why some users prefer it.

Is Riffusion free?

There is free access, with limits, and paid tiers above it. Availability through model aggregators also means you can try it inside a workflow you already have.

Can Riffusion generate vocals?

Yes, the current platform generates full songs with vocals rather than the instrumental clips the original project produced.

Why does Riffusion sound different from Suno?

Its lineage shows. Output has a recognisable character that does not sit in the homogenised middle of the category — which is a reason to choose it or avoid it depending on what you want.

Is Riffusion open source?

The original spectrogram-diffusion project was published openly and remains one of the best-documented examples of how these systems work. The current platform is a separate, more capable product.

Keep exploring