Chat-Based MIDI Generation
Generate MIDI from plain-language prompts: multi-turn refinement, up to three MIDI references, creative intent parsing, and provider handoff mid-conversation.
DocumentationChat-Based Generation
Core Feature
Chat-Based MIDI Generation
Aether's primary interface is a conversational chat. Describe what you want in plain English and Aether generates the MIDI. You can be as vague or specific as you like, from "a dreamy lo-fi chord progression" to "Cmaj7 - Dm7 - G7 - Cmaj7 in 4/4, 80 BPM, 8 bars with arpeggiated right hand."
Multi-Turn Conversations
Each message carries the recent conversation with it, so the AI understands context from previous generations. Aether sends a rolling window of the last several exchanges rather than the entire chat, which keeps requests fast and affordable on long sessions. In practice this means refinements chain naturally, while a detail from very early in a long conversation may need restating. This enables powerful iterative workflows:
- ●"Make a jazz chord progression in Bb major" → generates initial MIDI
- ●"Add a walking bassline" → layers bass over the existing chords
- ●"Make bar 5-8 more tense" → refines a specific section
- ●"Now transpose the whole thing down to Ab" → modifies the entire piece
MIDI References
Drop up to 3 MIDI files into the chat as reference material. Use them to combine, blend, remix, or provide context for the AI. The AI receives the full note data of your references and can work with them directly. A multi-track file keeps its parts named and still counts as one reference, so a four-part arrangement does not use up three slots. Aether tells you which files did not fit when you drop more.
- ●"Combine these two patterns into one": merges the attached references
- ●"Add a melody over this chord progression": generates new content that fits the first reference
- ●"Make this feel more like a bossa nova": restyles a reference
Importing Your Own MIDI covers this in full: the four ways to get a file in, what references do and do not carry over, and the editing controls that appear once a clip is loaded.
A clip with no tempo takes your session tempo
Many DAWs write no tempo into an exported MIDI clip; Ableton's "Export MIDI Clip" is the common case. Aether reads such a file at your current session tempo rather than guessing, so a four-bar clip stays four bars instead of arriving stretched. If a reference does carry its own tempo, that is used instead. Set the BPM you want before dropping the file in and the bar count will come through correctly.Seeing the AI Think
When a model reasons before it answers, Aether shows that reasoning in the chat as a collapsible Thought panel above the result. In Pro Mode the same panel holds the composition plan: the structure, section lengths, and part assignments the model settled on before writing a note. It is a useful way to tell a generation that misunderstood you from one that understood you and made a choice you disagree with.
The panel is on by default and controlled by Show AI reasoning in chat in Settings. Turning it off hides the panel; it does not change what is generated. Pro Mode & Batch shows what a composition plan looks like.
Reasoning is a description of musical decisions only. Aether asks the model to explain its thinking in those terms and nothing else, so the panel never exposes Aether's internal instructions.
Creative Intent Parsing
Aether extracts creative intent directly from your natural language. You don't need to manually select every parameter; the AI understands keywords and phrases:
| You Say | Aether Understands |
|---|---|
| "just a melody" | Melody only: no chords or bass |
| "give me a bassline" | Bass only texture |
| "chord progression" | Chords only: no melody |
| "verse and chorus" | Multi-section form with contrast |
| "add layers to this" | Accompaniment over existing reference |
| "rework this" | Reimagine existing MIDI material |
Continuity
Provider Handoff
You can change AI providers at any point during a conversation without losing context. When you switch, Aether summarizes the conversation so far and automatically attaches that summary to your next message, so the new provider has the essentials of what's been generated.
A system bubble confirms the handoff: "Provider switched. Previous conversation context will be carried over." The summary is one-shot: it's injected into your next send and then cleared.
Useful when a different provider is better at a specific task. For example, starting a melody with Claude and switching to Grok to add a bassline and chords while keeping the key, tempo, and feel consistent.