LLM-composed piano beats text-to-music models in 4-way AI composition comparison
UnfortunateHurricane · reddit · 2026-10-20
A developer without music background compared AI instrumental generation for game scoring: text-to-music models (YuE2, MiniMax-Music3, ACE-Step 1.5 XL) vs direct composition via paid frontier LLMs, with prompts and API costs included (each generation used 2-5% of weekly quota). Blind listening supported.
Findings:
- For solo piano, LLM-composed pieces are surprisingly good; text-to-music models are mediocre
- For orchestral pieces, text-to-music models fare better
- The author plans to invest more in the direct LLM composition route
All audio, prompts and costs are open-sourced on GitHub for reproduction.
More from Multimodal
- The Information: OpenAI has a plan to revive its Sora video app — shaunralston · 2026-10-20
- Creator says Claude via MCP replaced his video editing, finishing a full music video — AIandDesign · 2026-10-20
- Editor says Grok handled 90% of a music video edit — shares her 3 workflow tips — chaitu · 2026-10-20
- Creator generates tense cliff-road car clip in 30 seconds with Seedance 2.5, prompt included — LudovicCreator · 2026-10-20
- MOSS-Image ecosystem roundup: layout editors, line-art colorizer, color-depth control — optimisticalish · 2026-10-20
- Kling AI 4.0 debuts with 30-second clips, stereo audio, but doesn't outpace Wan 3.0 and Seedance 2.5 — The Decoder · 2026-10-20