Audio workbench
Record, trim, cut, copy, paste, undo and redo on an edit decision list — the source audio is never overwritten in place.
Studios
Four studios that share one library.
Audio, visual, radio and model training are separate crafts, so they get separate workspaces — but they operate on the same tracks, the same versions, the same rights and the same budget.
- Stem separation
- Stem workbench
- Mastering
- Quality lab
- Cover art
- Video editor
- AI DJ
- Smart radio
- LoRA training
- Model arena
Audio studio
From raw render to release master.
Editing and mastering are non-destructive follow-up versions, so improving a mix never costs you the take it came from.
Effects and analysis
Fade, gain, normalize, compression, EQ and a spectrum view, plus reverb, delay and distortion with keyboard shortcuts for the operations you repeat.
Stem separation
Split a mix into its track classes and get a real starting point for remixing, comping, video work or a DAW handoff.
Stem workbench
Solo, mute, group, gain, pan and fade across separated stems, and commit the result as a new version of the track.
Mastering profiles
Apply loudness and mastering profiles targeted at the destination — streaming, video or archive — rather than one generic loudness push.
Quality lab
Automatic findings for clipping, true peak, LUFS, silence, BPM and key, reported as issues you can act on before release.
Visual studio
Artwork and video that belong to the track.
Covers and videos are generated as their own jobs against the same project, with their own rights record and their own version history.
Cover generation
Automatic cover art runs in parallel with the audio job, with prompt, model, seed, size, format and variant count under your control.
Cover workflow
Regenerate, upload your own, crop, compare candidates and assign the chosen one — then embed it into the track metadata on export.
Video layer editor
A WYSIWYG editor over a scene manifest: layers, timing, backgrounds and effects, previewed on a timeline before anything is rendered.
Visualizers and formats
Ten visualizer presets plus custom and brand templates, rendered to 16:9, 9:16, 1:1 or a free profile for each platform you post to.
Lyrics on screen
Lines, scroll and karaoke styling driven by the same LRC timing the audio export uses, so the words stay in sync everywhere.
Durable rendering
Final renders run as durable FFmpeg jobs with history, templates and branding, then publish or share straight from the project.
DJ and radio
Music that keeps playing, under a budget you set.
Continuous generation is genuinely useful and genuinely expensive, so every part of it runs inside hard limits you control.
AI DJ conversations
A conversational surface for generating, adjusting and saving music, with named conversations, history, auto-titling and per-conversation settings.
Stations
Start from a preset, build a custom station, or spin one up from a song you already like — then edit, export or delete it.
Continuous generation
The queue refills itself while you listen, with auto-save, playlist capture and a now-playing view that keeps the history.
Hard stop rules
Track count, total duration, cost ceiling and idle timeout are enforced platform-side. A radio session cannot quietly run up a bill overnight.
Ambience
Optional vinyl, static and brown-noise beds are a local playback effect — never generated audio and never part of your export.
Creator copilot
A contextual assistant that can act inside the product — propose a change, prepare a job, adjust metadata — with every action reviewable.
Models and training
Your own sound, trained and versioned.
Model and adapter work is part of the platform, with provenance, licence and quota tracked as strictly as anything else you generate.
Model catalog
Status, size, licence, version and capability for every available model, so a choice is informed rather than a guess at a dropdown label.
Adapters and LoRAs
Add LoRA and LoKr adapters with metadata, licence, activation and scale; stack several, version them and scope who may use them.
Datasets
Upload or import training material, scan and edit it, preview items and delete what should not be there — with the rights position recorded per item.
Preprocessing
Auto-labelling, format conversion, correction and reusable presets so a dataset becomes trainable without leaving the product.
Training runs
Start, stop and resume runs with full parameter control, and follow loss curves, logs, checkpoints and audio samples as they arrive.
Model arena
Compare models and adapters blind on the same prompt and evaluate the result, instead of trusting an impression from a single lucky render.
The short version
- Quality lab
- Clipping, true peak, LUFS, silence, BPM and key findings
- Video profiles
- 16:9, 9:16, 1:1 and free, with ten visualizer presets
- Radio limits
- Track, duration, cost and idle stops enforced platform-side
- Training
- LoRA and LoKr with checkpoints, samples, merge and blind arena