A Map of the Tools
AI music is not one big tool but a chain of small ones, each owning one stage.
Related
The problem
"Making music with AI" sounds like one task but is a chain: decide what you want, describe it, generate material, split parts, fix details, mix, master, deliver. Each stage has dedicated tools, and there are all-in-one tools that do every stage adequately but none well. A map comes before a choice.
The map
Four groups by input and output. Text to audio takes a description and returns a finished track or a fragment. Audio to audio takes reference audio for style transfer, continuation or separation. Symbolic tools work on MIDI and scores, offering the most control but relying on sample libraries for sound. Assistants handle stem separation, tuning, mastering and sound effects, saving time without composing. Most real workflows mix several groups rather than staying in one.
Choosing
Ask what the deliverable is first: a finished song, a cue, editable stems, or a handful of sound effects. A finished song suits text to audio plus manual polish; anything going back into a DAW needs stems or MIDI; film and game work needs timecode alignment. Then ask how much control you need: is a slot-machine generation acceptable, or must harmony and orchestration be specified. Finally ask about rights: can the output be used commercially, and is the training data accounted for.
What to listen for
In the single-line folk tune, what melody tools find easy; in the piano rondo, how complex the texture gets; in the drum loop, whether the groove holds. A tool's limits show first in dense textures and long forms.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Sources
- 生成式音乐工具的功能分类，依公开产品文档归纳，未抄文本
- 音频与 MIDI 工作流的阶段划分，依制作通识核对，未抄原文
Used for fact-checking only; all prose is original.