Suno launched Visuals on November 18, extending its product from music generation into video. The unit of output is a 15-second clip; the workflow is stitching several into a full music video.
The design choice
Fifteen seconds is short for a music video and long for a generation. That is deliberate: it matches the length a current video model can hold coherence for, and it lets the user keep the parts that work.
A single long generation is a slot machine — one bad frame ruins the take, and you regenerate the whole thing. Clips are composable. You generate six, keep four, regenerate two. That is a much better editing model for a generative tool, and it is the same reason image generation moved to inpainting rather than whole-image regeneration.
The model is not Suno's
Reporting says the underlying video model is based on MiniMax. That is worth stating plainly because it tells you where Suno's advantage actually is.
Suno's asset is the audio model and the workflow: it knows the song structure, the beat grid, the lyrics and the mood. Pairing that with a rented video model is a sensible division of labour — video generation is capital-intensive and improving monthly, so building it in-house would be a poor use of the round.
The pricing signal
Video generation is billed separately from song credits, which is the honest way to price it. Video inference costs orders of magnitude more than audio inference per second of output, and bundling them would either make music expensive or make video unprofitable.
Separate credits also let Suno throttle the expensive modality without touching the cheap one. Expect every generative product to converge on this: one subscription, metered by modality.
What it means competitively
The generative media market is consolidating around suites. A music tool that cannot produce a video is competing with tools that can, and the switching cost for a creator is mostly habit rather than technology.
Suno's risk is the opposite: by adding video it competes with dedicated video tools that are better at video, while its music advantage matters less in a music video than the visuals do.
The rights question, again
Music videos are the format where rights get complicated fastest. A generated video of a performer who resembles a real artist, set to a generated song that resembles a real one, is a legal question with no settled answer.
Suno has been through this once already on the audio side. The video side has more parties: the model provider, the training data behind it, and the likeness of anyone the video depicts.
The practical read
For independent musicians, this is the first genuinely useful tool in the category: a lyric video or a visualiser used to cost a few hundred dollars and a week. Fifteen-second clips are enough for a social cut, which is where most of the demand actually is.
For anyone building in the space, the lesson is the composability one. Generative tools that expose intermediate artefacts — clips, stems, layers — get adopted faster than tools that only return a finished file, because creators need to fix things, not just accept them.