iZotope has released Photon, an audio tool combining stem separation with restoration features, including options to clean up vocals and reduce background noise. It runs locally on a GPU, giving audio creators a new set of machine-learning tools without making cloud processing part of the stated setup.
Watch Desk analysis
What happened
Photon is available as a standalone app and an ARA plug-in. DJ Mag reports that its 12 custom-trained machine-learning models are split between Demix, for separating audio, and Denoise, for restoration.
Among the listed tools are a Voice Separator that can separate up to six speakers on one track, Voice Restore for vocal clean-up, and a drum model that breaks a performance into kick, snare, hi-hat, toms and cymbals. A Cinema feature separates video or audio into dialogue, music, sound effects and other material. The restoration tools are designed to tackle noise such as wind, traffic and indoor room sound.
The report says Photon works on macOS 14 or later and Windows 10 or later. It lists a discounted price of $259 until 7 December, rising to $349 afterwards. Those are US-dollar prices; the supplied details do not establish local pricing elsewhere. DJ Mag’s report on Photon describes the product and its features.
Why it matters
The appeal is the combination: separating sources, restoring rough recordings and handling particular tasks such as drum breakdowns in one locally run tool. For producers, editors and people working with dialogue, that could mean fewer trips between specialist tools. The practical value will depend on how well each model handles real, messy recordings, not just how tidy its feature list looks.
Our read
This is a specific AI-computing development with clear creative uses, rather than a vague promise to make audio “smarter”. The local GPU option and named models give prospective users something concrete to assess. If you work with audio, check operating-system compatibility and the price that applies where you are, then judge the results on your own material. No machine-learning model has yet persuaded a traffic jam to become a studio.
What to watch
- Whether creators report consistent separation and restoration quality on varied recordings.
- Which GPUs and ARA-compatible hosts work well in practice.
- Whether iZotope adds models, updates the tool or publishes more detail on performance.
Discussion spark: Would you rather have one audio tool covering separation and restoration, or keep using specialist tools for each job?
Sources and evidence
- Source update (17 November 2026, 18:01 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.