The root of the limits
The system learned what music tends to sound like, not what a piece is supposed to do. Long-form coherence, precise emotional intent, and exact consistency all require holding a goal across the whole work, which that kind of learning does not provide.
Three limits and what each requires
Long-form coherence
- Requires later sections to relate to earlier ones by plan
- Pattern-based generation drifts because nothing rewards a global plan
Emotional intent
- Requires expressing a specific feeling you have in mind
- Recognizable moods are patterns in the data; your private intent is not
Consistency
- Requires a stable reference held across the whole track
- Pattern-based generation tends to vary, which helps variation and hurts consistency
Originality and rights
Because the training material is existing music, questions arise about whether output is genuinely new, whether it carries the influence of specific works, and who holds rights over the result. These are open questions rather than settled rules, and they vary by context. This is a description of the questions, not legal advice.