ComfyUI Extension: JK AceStep Nodes

Authored by jeankassio

Created

Updated

8 stars

Advanced nodes optimized for ACE-Step audio generation in ComfyUI.

Custom Nodes (0)

    README

    JK AceStep Nodes for ComfyUI

    Advanced nodes optimized for ACE-Step audio generation in ComfyUI.

    šŸ“¦ Installation

    cd ComfyUI/custom_nodes/
git clone https://github.com/jeankassio/JK-AceStep-Nodes.git

    Restart ComfyUI. Nodes will appear under JK AceStep Nodes/ categories.

    šŸŽµ Nodes

    Ace-Step KSampler (Basic)

    Full-featured sampler with quality check, advanced guidance (APG, CFG++, Dynamic CFG), anti-autotune smoothing, and automatic step optimization.

    Category: JK AceStep Nodes/Sampling

    Ace-Step KSampler (Advanced)

    Extended sampler with start/end step control for multi-pass workflows and refinement.

    Category: JK AceStep Nodes/Sampling

    Ace-Step Prompt Gen

    Prompt generator with 150+ professional music styles (Electronic, Rock, Jazz, Classical, Brazilian, World Music, and more).

    Category: JK AceStep Nodes/Prompt

    šŸŽ¤ Lyrics Generators

    Ten AI-powered lyrics generation nodes supporting various LLM providers:

    Ace-Step OpenAI Lyrics

    Lyrics generation using OpenAI GPT models.

    Supported Models (December 2025):

    • gpt-5.1 - Reasoning model (latest)
    • gpt-5.1-codex - Coding optimized
    • gpt-5 - High performance
    • gpt-5-pro - Professional variant
    • gpt-4o - Multimodal (recommended)
    • gpt-4o-mini - Fast variant
    • gpt-4-turbo - High performance
    • gpt-4 - Stable base
    • o3 - Reasoning model
    • o3-mini - Compact reasoning
    • o1 - Advanced reasoning
    • o1-mini - Compact advanced reasoning

    Category: JK AceStep Nodes/Lyrics

    Ace-Step Claude Lyrics

    Lyrics generation using Anthropic Claude models.

    Supported Models (December 2025):

    • claude-opus-4.5 - Latest flagship (recommended)
    • claude-opus-4.1 - Previous flagship
    • claude-sonnet-4.5 - Latest balanced
    • claude-sonnet-4 - Previous balanced
    • claude-haiku-4.5 - Latest fast
    • claude-haiku-3.5 - Previous fast
    • claude-3-5-sonnet-20241022 - Snapshot variant
    • claude-3-5-haiku-20241022 - Snapshot variant
    • claude-3-opus-20250219 - Dated variant

    Category: JK AceStep Nodes/Lyrics

    Ace-Step Gemini Lyrics

    Lyrics generation using Google Gemini API.

    Supported Models (December 2025):

    • gemini-3-pro - Latest pro model (recommended)
    • gemini-2.5-flash - Fast with latest capabilities
    • gemini-2.5-flash-lite - Ultra-fast variant
    • gemini-2.5-pro - High quality
    • gemini-2.0-flash - Previous generation
    • gemini-2.0-flash-lite - Previous generation lite

    Category: JK AceStep Nodes/Lyrics

    Ace-Step Groq Lyrics

    High-speed lyrics generation using Groq API.

    Supported Models (December 2025):

    • llama-3.3-70b-versatile - Meta Llama 3.3 70B (best quality)
    • llama-3.1-8b-instant - Meta Llama 3.1 8B (fast)
    • llama-guard-4-12b - Meta Guard model
    • deepseek-v3 - DeepSeek V3
    • mistral-small-3 - Mistral Small v3
    • gpt-oss-120b - OpenAI OSS 120B
    • gpt-oss-20b - OpenAI OSS 20B
    • Plus additional production models

    Category: JK AceStep Nodes/Lyrics

    Ace-Step Perplexity Lyrics

    Lyrics generation using Perplexity Sonar models.

    Supported Models (December 2025):

    • sonar - Standard model
    • sonar-pro - Professional variant
    • sonar-reasoning - Reasoning-focused
    • sonar-reasoning-pro - Advanced reasoning
    • sonar-deep-research - Deep research variant

    Category: JK AceStep Nodes/Lyrics

    Ace-Step Cohere Lyrics

    Lyrics generation using Cohere Command models.

    Supported Models (December 2025):

    • command-a-03-2025 - Latest Command A
    • command-r7b-12-2024 - December 2024 variant
    • command-r-plus-08-2024 - R+ August 2024
    • command-r-08-2024 - R August 2024
    • command-a-translate - Translation specialist
    • command-a-reasoning - Reasoning-focused
    • command-a-vision - Vision capabilities
    • aya-expanse-32b - Aya Expanse 32B
    • aya-expanse-8b - Aya Expanse 8B
    • aya-vision - Aya with vision
    • aya-translate - Aya translation specialist

    Category: JK AceStep Nodes/Lyrics

    Ace-Step Replicate Lyrics

    Lyrics generation using Replicate API models.

    Supported Models (December 2025):

    • meta/llama-3.1-405b-instruct - 405B instruction-tuned
    • meta/llama-3.1-70b-instruct - 70B instruction-tuned
    • meta/llama-3.1-8b-instruct - 8B instruction-tuned
    • meta/llama-3-70b-instruct - Llama 3 70B
    • meta/llama-2-70b-chat - Llama 2 chat 70B
    • mistralai/mistral-7b-instruct-v0.3 - Mistral 7B v0.3
    • mistralai/mistral-small-24b-instruct-2501 - Mistral Small 24B
    • mistralai/mixtral-8x7b-instruct-v0.1 - Mixtral MoE

    Category: JK AceStep Nodes/Lyrics

    Ace-Step HuggingFace Lyrics

    Lyrics generation using HuggingFace Inference API.

    Supported Models (December 2025):

    • meta-llama/Llama-3.1-405B-Instruct - Large instruction-tuned
    • meta-llama/Llama-3.3-70B-Instruct-Turbo - Llama 3.3 70B turbo
    • meta-llama/Llama-3.1-70B-Instruct - 70B instruction-tuned
    • mistralai/Mistral-Large - Large Mistral variant
    • microsoft/Phi-4 - Phi-4 model
    • deepseek-ai/deepseek-v3 - DeepSeek V3
    • Qwen/Qwen2.5-72B-Instruct - Qwen 2.5 72B
    • google/gemma-2-27b - Gemma 2 27B
    • tiiuae/falcon-180b - Falcon 180B

    Category: JK AceStep Nodes/Lyrics

    Ace-Step Together AI Lyrics

    Lyrics generation using Together AI serverless models.

    Supported Models (December 2025):

    • meta-llama/Llama-3.3-70B-Instruct-Turbo - Llama 3.3 70B turbo
    • meta-llama/Llama-3.1-405B-Instruct-Turbo - Llama 3.1 405B turbo
    • mistralai/Mistral-Small-24B-Instruct-2501 - Mistral Small 24B
    • Qwen/Qwen2.5-72B-Instruct - Qwen 2.5 72B
    • deepseek-ai/DeepSeek-V3 - DeepSeek V3
    • moonshotai/Kimi-K2-Instruct - Kimi K2
    • GLM-4-Plus - GLM 4 Plus
    • Nous-Hermes-3-70B - Nous Hermes 3 70B
    • Plus 100+ additional models available

    Category: JK AceStep Nodes/Lyrics

    Ace-Step Fireworks Lyrics

    Lyrics generation using Fireworks AI models (100+ available).

    Supported Models (December 2025):

    • deepseek-ai/deepseek-v3 - DeepSeek V3 (latest)
    • deepseek-ai/deepseek-r1 - DeepSeek R1 reasoning
    • Qwen/Qwen3-235B-A22B-Instruct - Qwen 3 235B
    • Qwen/Qwen2.5-72B-Instruct-Turbo - Qwen 2.5 72B turbo
    • meta-llama/Llama-4-Maverick-17B - Llama 4 Maverick
    • meta-llama/Llama-4-Scout-17B - Llama 4 Scout
    • meta-llama/Llama-3.3-70B-Instruct - Llama 3.3 70B
    • meta-llama/Llama-3.1-405B-Instruct - Llama 3.1 405B
    • mistralai/Mistral-Large-3-675B-Instruct - Mistral Large 675B
    • mistralai/Mistral-Small-24B-Instruct-2501 - Mistral Small 24B
    • google/GLM-4.6 - GLM 4.6
    • moonshotai/Kimi-K2 - Kimi K2
    • google/Gemma-3-27b - Gemma 3 27B
    • Plus 90+ additional models available

    Category: JK AceStep Nodes/Lyrics

    Ace-Step Save Text

    Text saver with auto-incremented filenames and folder support. Works with any lyrics generator.

    Category: JK AceStep Nodes/IO

    šŸŽØ JKASS Custom Sampler

    Just Keep Audio Sampling Simple - custom sampler optimized for audio generation.

    Available Variants

    • jkass_quality - Second-order Heun method for maximum audio quality

      • Superior accuracy and detail preservation
      • Recommended for final renders
      • ~2x slower than fast variant

    • jkass_fast - First-order Euler method for faster generation

      • Optimized for speed with vectorized operations
      • Good quality with reduced computation time
      • Best for iterations and previews

    Key Features

    • No noise normalization (preserves audio dynamics)
    • Clean sampling path (prevents word cutting/stuttering)
    • Optimized for long-form audio

    Select your preferred variant from any sampler dropdown (default: jkass_quality).

    šŸ“Š Recommended Settings

    • Sampler: jkass_quality (for best quality) or jkass_fast (for speed)
    • Scheduler: sgm_uniform
    • Steps: 80-100
    • CFG: 4.0-4.5
    • Anti-Autotune: 0.25-0.35 (vocals), 0.0-0.15 (instruments)

    āœ… Extra settings for reducing 'AI' female vocals

    • Sampler: jkass_quality (for best quality) or jkass_fast (for speed)
    • Frequency Damping: 0.15-0.5 for female vocals to reduce metallic sizzle (0=disabled)
    • Temporal Smoothing: 0.02-0.12 to reduce pitch quantization and temporal discontinuities
    • Beat Stability: 0.05-0.2 to preserve stable rhythmic strikes and avoid per-frame jitter
    • Anti-Autotune: 0.25-0.35 (vocals), 0.0-0.15 (instruments)

    šŸŽÆ Quality Check Feature

    Automatically tests multiple step counts to find optimal settings for your prompt.

    Important: Quality scores are comparative only. Compare within same prompt/style. Electronic music naturally scores lower than acoustic (both can be excellent quality).

    šŸ”§ Troubleshooting

    • Word cutting/stuttering: Use jkass_quality sampler, disable advanced optimizations
    • Metallic voice: Increase anti_autotune_strength to 0.3-0.4
    • AI-sounding female voice: Try the following sequence:
      1. Use jkass_quality and 80-120 steps, CFG 4.0-4.5, APG enabled
      2. Set Anti-Autotune (0.25-0.35), Frequency Damping (0.15-0.4), Temporal Smoothing (0.02-0.06)
      3. Use the Prompt Gen with voice_style -> natural_female and add 'breathy, micro pitch variation' in extra prompt
      4. Decode using a high-quality VAE/vocoder (HiFi-GAN, or validated VAE) for improved timbre
      5. If still metallic: de-esser and mild EQ cut at 7-12 kHz; add subtle formant correction and breath overlay
      Optional: Use the Ace-Step Post Process node to apply a quick de-essing (reduce 6-10 kHz energy), spectral smoothing, and subtle breath overlay to humanize the vocal further.
    • Poor quality: Increase steps (80-120), use CFG 4.0-4.5, enable APG, try jkass_quality sampler

    šŸ“ Project Structure

    JK-AceStep-Nodes/
ā”œā”€ā”€ __init__.py                    # Central node aggregator
ā”œā”€ā”€ ace_step_ksampler.py           # Samplers (Basic + Advanced)
ā”œā”€ā”€ ace_step_prompt_gen.py         # Prompt generator (150+ styles)
ā”œā”€ā”€ lyrics_nodes.py                # 10 lyrics generators consolidated
ā”œā”€ā”€ ace_step_save_text.py          # Text saver node
ā”œā”€ā”€ requirements.txt
ā””ā”€ā”€ py/
    ā””ā”€ā”€ jkass_sampler.py           # Custom audio sampler

    Available Lyrics Generators

    • OpenAI - gpt-4o, gpt-4-turbo, gpt-4, gpt-3.5-turbo, and more
    • Anthropic Claude - Claude 3.5 Sonnet, Claude 3 Opus/Sonnet/Haiku
    • Google Gemini - gemini-2.5-flash, gemini-2.5-pro, gemini-2.0-flash, gemini-1.5-pro/flash
    • Groq - Llama 3.3 70B, Llama 3.1 8B, Llama Guard 3, GPT-OSS (120B/20B), and Llama 4 preview models
    • Perplexity - Sonar, Sonar Pro, Sonar Reasoning (with 128k context)
    • Cohere - Command A/R+ (with reasoning & vision), Aya (multilingual)
    • Replicate - Llama 3.1 (405B/70B/8B), Mistral Small/Nemo, Mixtral
    • HuggingFace - Llama 3.1, Mistral, DeepSeek, Qwen, Falcon, and 100+ more
    • Together AI - Llama 3.3/3.1, DeepSeek, Qwen 3, Mistral variants, and 50+ more
    • Fireworks AI - DeepSeek V3/R1, Qwen 3, Llama 3.3/3.1, Mistral Large/Small, GLM, and 90+ more

    šŸ—£ļø How to Use the Vocoder (ADaMoSHiFiGAN)

    To enable audio conversion with the vocoder (for improved final audio quality):

    https://huggingface.co/ACE-Step/ACE-Step-v1-3.5B/tree/main/music_vocoder

    1. Obtain the vocoder files:

      • diffusion_pytorch_model.safetensors (vocoder model)
      • config.json (vocoder configuration)

    2. Place both files in the folder:

      • JK-AceStep-Nodes/vocoder/

      The final path should be:

      JK-AceStep-Nodes/vocoder/diffusion_pytorch_model.safetensors
JK-AceStep-Nodes/vocoder/config.json

    3. Done!

      • The system will automatically detect these files when using nodes with vocoder enabled.
      • If the files are not present, audio will be generated without the vocoder.

    Tip: Always use the correct file pair (model + config) to avoid artifacts or loading errors.

    šŸ“„ License

    MIT License

    Version: 2.3
    Last Updated: December 2025

    šŸŽµ Happy music generation!