Comfyui-Song-Generation-Suite

🎵 SongGenerationSuite – AI Song Writer for ComfyUI (ACE 1.5)

SongGenerationSuite is a ComfyUI custom node designed to generate complete, structured songs (lyrics + music prompt) and convert them into conditioning compatible with ACE 1.5 audio generation workflows.

It allows you to go from a simple idea to a fully conditioned song ready to be sampled with a KSampler, without writing complex prompts or lyrics manually.

---

✨ What this node does

From a single node, SongGenerationSuite generates:

• 🎼 A detailed music prompt (style, instruments, mood, structure)
• ✍️ Fully structured lyrics with strict bracket tags
  ([Intro], [Verse], [Chorus], [Bridge], [Outro], etc.)
• 🧠 Text conditioning compatible with ACE 1.5
• 🧪 Debug outputs (full LLM output & prompt used)

All outputs are designed to plug directly into an ACE 1.5 + KSampler pipeline.

---

🧩 Workflow overview

Minimal graph:

1. ACE 1.5 Checkpoint Loader
2. SongGenerationSuite
3. KSampler
4. ConditioningZeroOut
5. VAE Decode Audio / Save Audio

---

🔌 Inputs & Wiring (ACE 1.5)

CLIP input (required)

SongGenerationSuite does not load ACE by itself.

You must:

• Load ACE 1.5 using a Checkpoint Loader
• Connect the loader’s CLIP output to SongGenerationSuite → clip

This ensures:

• The generated conditioning uses the same text encoder
• Full compatibility with the ACE pipeline

ACE 1.5 Checkpoint Loader
 ├─ MODEL → KSampler (model)
 └─ CLIP  → SongGenerationSuite (clip)

---

Conditioning output

The conditioning output from SongGenerationSuite must be connected to:

• KSampler → positive
• ConditioningZeroOut → conditioning

This is the recommended ACE wiring.

---

🎛 User parameters

Core songwriting

• song_topic – Main theme or story of the song
• lead_singer_sex – Vocal profile used in the prompt
• music_style – Base genre
• music_style_custom – Optional detailed or hybrid style
• lyrics – Optional user lyrics
• strict_user_lyrics – Enforce lyrics exactly if enabled
• lyrics_language – Language of the lyrics

Musical structure

• bpm – Target tempo
• duration – Target song length (seconds)
• timesignature – Musical time signature
• keyscale – Musical key
• lyrics_strength – Conditioning strength

Generation

• seed – Reproducibility
• batch_size – Number of generations

LLM model & performance

• model – Built-in or custom HF model
• custom_model_repo_id – HuggingFace repo ID
• quantization – FP16 / 8-bit / 4-bit
• attention_mode – Attention backend
• use_torch_compile – Optional optimization
• device – auto / cuda / cpu
• keep_model_loaded – Keep model in VRAM

Decoding controls

• max_tokens – Maximum generated tokens
• temperature – Creativity level
• top_p – Nucleus sampling
• num_beams – Beam search
• repetition_penalty – Reduce repetition

---

📦 Installation

cd ComfyUI/custom_nodes
git clone https://github.com/YOURNAME/SongGenerationSuite.git
pip install -r requirements.txt

Restart ComfyUI.

---

📝 Notes

• Prompts are written in English for LLM stability
• Lyrics language is controlled independently
• Designed for ACE 1.5
• Ideal for music generation and creative pipelines

---

🚀 Use cases

• AI song prototyping
• Music prompt generation
• ACE 1.5 audio experiments
• Rapid creative iteration