Why is CFG disabled, and can I use negative prompts?

This workflow intentionally disables classifier-free guidance using ConditioningZeroOut for fast, stable sampling. Negative prompts are ignored in this setup. If you need negatives, replace ConditioningZeroOut with an explicit negative encoder path and enable CFG in KSampler—but that changes the workflow’s behavior.

How do I control duration and structure (e.g., 10–30 seconds)?

Set the length in EmptyAceStep1.5LatentAudio (seconds or samples, depending on the node’s fields). The model interprets structure via text (e.g., “16‑bar trap loop,” “slow build and soft drop”). For longer pieces, generate multiple segments and arrange them in a DAW.

What sampler settings should I use for best results?

Keep KSampler at 8 steps with ModelSamplingAuraFlow connected—this is tuned for ACE-Step 1.5 XL Turbo. If you hear artifacts, try modest increases (e.g., 10–12 steps). Maintain CFG disabled to match the workflow’s design.

What format does it export, and can I change it?

The graph saves MP3 via SaveAudioMP3. You can adjust its bitrate if the node exposes that option. If you prefer WAV, swap in a compatible WAV saver node and connect it after VAEDecodeAudio.

ACE-Step 1.5XL Turbo: Text to Music

This workflow turns a plain-language text prompt into finished music using the ACE-Step 1.5 XL Turbo model. Your prompt is encoded by TextEncodeAceStepAudio1.5 (backed by DualCLIPLoader) into audio-aware conditioning. EmptyAceStep1.5LatentAudio creates a blank latent audio canvas at your chosen length, and UNETLoader loads the distilled 4B ACE-Step checkpoint (acestep_v1.5_xl_turbo_bf16.safetensors). ModelSamplingAuraFlow configures the sampler schedule tuned for ACE-Step, and KSampler performs just 8 diffusion steps to synthesize the latent audio. The audio is then reconstructed with VAELoader + VAEDecodeAudio and saved via SaveAudioMP3.

What makes this workflow practical is its “turbo” configuration: no classifier-free guidance (CFG) is used, enabled by ConditioningZeroOut feeding an intentionally empty negative conditioning. This keeps prompts straightforward (focus only on what you want), while ModelSamplingAuraFlow and KSampler deliver fast, stable results. The graph is organized into Model and Prompt groups for clarity, and uses PrimitiveInt/PrimitiveNode for quick edits to seed, steps, and duration. The result is a tight, teachable text-to-music pipeline you can run repeatedly for variations or integrate as a reusable subgraph.

API

Use this workflow from code

Every Comfy workflow is a JSON graph. The payload below is this workflow, exactly as ComfyUI runs it — fetch it from the URL, keep it in version control, or load it in ComfyUI and run it node by node.

GET https://comfy.org/workflows/download/9851c174a194.json

Fetching workflow JSON…

Run it from Python or TypeScript with the Comfy SDK. The same code targets Comfy Cloud or a ComfyUI you host yourself — only the base URL changes.

# Install (beta)
pip install comfy-sdk        # Python
npm i @comfyorg/sdk          # TypeScript

# Run "ACE-Step 1.5XL Turbo: Text to Music" (Python)
from comfy_sdk import Comfy

client = Comfy(api_key="comfyui-...")

# This workflow, exported in API format (see note below)
wf = client.workflows.from_file("audio_ace_step1_5_xl_turbo_api.json")

job = client.run(wf)
for output in job.get_outputs("107"):  # SaveAudioMP3
    output.to_file(output.name)

The SDK takes a workflow in API format: open this workflow in ComfyUI and use File → Export Workflow (API). API access requires a Comfy Cloud plan with an API key. SDK docs · Get an API key

FAQ

Frequently Asked Questions

View all workflows