iZotope Photon Introduces GPU-Powered AI Stem Separation and Audio Restoration
iZotope has released Photon, an AI-powered audio processing application that combines music stem separation, drum extraction, speaker isolation, and dialogue restoration. Developed in collaboration with Boris FX, Photon is available as a standalone application and an ARA2 plug-in. Its machine-learning models run locally on the user’s GPU.
Photon uses 12 custom-trained machine-learning models organized into two categories: Demix for audio separation and Denoise for noise reduction and audio restoration. The models are designed for specific sources and recording conditions, including music, drums, speech, and environmental noise.
Music and Drum Stem Separation
Photon can split a finished music mix into four stems: vocals, drums, bass, and other instruments. It offers two models for this task: Music 4 Stems for general separation and Music 4 Stems Ultra for more precise results when individual stems need to stand on their own.
The Drums 5 Stems model separates a drum recording into kick, snare, hi-hat, toms, and cymbals. It works with both drum buses and full mixes, giving producers and mixing engineers access to individual drum elements even when the original multitrack session is unavailable.
Photon also includes dedicated voice and dialogue separation models. Voice 2 Stems splits a recording into voice and music, while Voice Separator can isolate up to six speakers into separate stems, along with room tone and background noise. Cinema 4 Stems separates audio or video soundtracks into voice, music, sound effects, and other elements for post-production workflows.
AI Restoration and DAW Integration
Photon also includes models for cleaning up dialogue and vocal recordings. Voice Restore reduces noise and echo, adjusts recording levels, and improves microphone tone. The Indoor and Indoor Noisy models target room noise and reverberation, while the Outdoor models address general background noise, traffic, and wind.
The ARA2 plug-in brings Photon’s processing engine directly into compatible DAWs, including Pro Tools, REAPER, Cubase, Sequoia, and Samplitude. Because ARA2 gives Photon access to the full audio clip, users can run multiple models, switch between them, and compare results without waiting for the entire file to be processed again.
All 12 machine-learning models run locally on the user’s GPU, so audio files do not need to be uploaded to cloud servers. Photon also supports preview processing from any point in a recording, allowing users to hear results within seconds without processing the entire file first.
System Requirements and Pricing
Photon runs on Windows 10 or later and macOS 14 or later. Windows users need a dedicated GPU with at least 4GB of VRAM. Mac users need an Apple Silicon processor (M1 or later); Intel-based Macs are not supported. Photon is available as a standalone application and an ARA2 plug-in, with no native VST3 or AU versions.
The introductory price is $259, reduced from the regular $349. This offer is available through December 7, 2026. iZotope also offers a 14-day free trial.
Photon combines music and drum stem separation with dialogue restoration in a single GPU-powered application. For music producers and mixing engineers, its practical value will depend largely on the quality of the extracted stems and how well the results fit into existing production workflows.
