ASLP Lab: DiffRhythm

DiffRhythm is a diffusion-based lyrics-to-song model that creates vocals and accompaniment together. Structured lyrics establish the song, while a style description and optional reference audio guide its musical character. Its official model page describes a ninety-five-second standard mode and a 285-second extended mode, allowing either a shorter demo or a more developed composition. Verse and chorus sections help communicate the intended song layout. Use cases: lyric demos, reference-guided songwriting prototypes and full-song soundtrack sketches built around an existing set of words.

模型组织

aslp-lab

上下文长度

未提供

输入

文本 + 音频

输出

音频

curl "$ROUTER_BASE_URL/api/v1/chat/completions" \
  -H "Authorization: Bearer $ROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "fal/fal-ai/diffrhythm",
  "messages": [
    {
      "role": "user",
      "content": "Gentle acoustic music with warm piano and soft guitar"
    }
  ],
  "modalities": [
    "text",
    "audio"
  ],
  "stream": true,
  "audio": {
    "format": "wav",
    "duration": 95,
    "lyrics": "[verse]\nMorning light across the bay\n[chorus]\nWe greet a bright new day"
  }
}'