Node Development Guidelines

Input/Output Types

Match ComfyUI conventions:

• MODEL, VAE, CLIP - Standard model types
• CONDITIONING - For conditioned generation
• LATENT - For latent audio
• AUDIO - For waveform audio
• STRING - For text (prompts, lyrics, metadata)
• INT, FLOAT, BOOLEAN - For parameters

Metadata Formatting

Use YAML format (as ace15.py expects):

meta = 'bpm: {}\\nduration: {}\\nkeyscale: {}\\ntimesignature: {}'.format(
    bpm, duration, keyscale, timesignature
)

Audio Code Format

Audio codes are integers offset from audio_start_id (151669):

# Generated codes are relative (0-based from audio range)
# ace15.py handles the offset internally
audio_codes = [1234, 5678, 910]  # These map to 152903, 157347, 152579

LM Generation Parameters

Exposed in sample_manual_loop_no_classes():

• cfg_scale: Float (default 2.0)
• temperature: Float (default 0.85)
• top_p: Float (default 0.9)
• top_k: Int (default None)
• seed: Int (default 1)
• min_tokens: Int (duration * 5 for 5Hz)
• max_new_tokens: Int (same as min_tokens for exact length)

---

Questions to Ask Before Building

1. Does a similar node exists

   • If yes → Reuse or extend it

2. Does ComfyUI core provide this?

   • If yes → Use the core functionality

3. Can this be done with existing nodes connected together?

   • If yes → Create a workflow example instead

4. Is this feature used by >50% of users?

   • If no → Consider making it optional/extensible

5. Can I pass this as kwargs?

   • If yes → Create a simple wrapper node

---

Code Style & Standards

Follow Existing Patterns

Node Template

class AceStepNodeName:
    @classmethod
    def INPUT_TYPES(cls):
        return {
            "required": {
                "param": ("TYPE", {"default": value}),
            },
            "optional": {
                "optional_param": ("TYPE",),
            }
        }
    
    RETURN_TYPES = ("TYPE1", "TYPE2")
    RETURN_NAMES = ("output1", "output2")
    FUNCTION = "execute"
    CATEGORY = "Scromfy/Ace-Step/"
    
    def execute(self, param, optional_param=None):
        # Implementation
        return (output1, output2)

Documentation

Every node should have:

• Clear docstring explaining purpose
• Input parameter descriptions
• Output descriptions
• Example usage in comments