ACE-Step v1.5 XL Base

ACE-Step v1.5 XL Base is the bigger sibling of the plain ACE-Step node in this pack, and the difference is in the name: "XL" and the extra scheduler controls. Where the base node hides its diffusion internals, this one hands you the sampler interval and a modular-scheduler escape hatch via JSON. It's the same music-generation family - prompt, lyrics, key, BPM, time signature - but built for people who've decided the defaults aren't good enough and want knobs to turn.

The model itself is from the ACE-Step project (the family behind the 3.5B checkpoint that impressed people in 2025 for being roughly 15x faster than LLM-based music generators). The v1.5 XL tier is the API's bigger-capacity offering, so as with everything in this pack there are no weights on your disk - you're renting a cloud generation.

What you set

positivePrompt is required. Everything from the base node carries over: settings.lyrics, settings.keyScale, settings.bpm, settings.timeSignature, settings.vocalLanguage, settings.guidanceType (apg/cfg), steps (100), CFGScale (10), duration (30–300s), plus the audio + repaintingStart/repaintingEnd editing path and strength.

The XL-specific extras:

• settings.cfgIntervalStart / settings.cfgIntervalEnd - where in the denoising schedule guidance kicks in and stops, as a fraction (0.0 = first step, 1.0 = last). Shortening the interval is a legit trick to avoid overcooked high-frequency detail at the end of a track.
• advancedFeatures - a gated toggle whose advanced_json accepts JSON for advancedFeatures.modularScheduler. That's the power-user path: a modular scheduler lets you mix scheduler behavior across the diffusion schedule. If you don't know what you'd put there, leave it off - the field exists precisely so power users can shove custom config in without a node update.

Everything else is shared with the base node: outputFormat MP3/WAV/FLAC/OGG, ttl, and the output audio (AUDIO dict, waveform + sample rate) → Save Audio.

Installing

This is one node in the Runware/ComfyUI-Runware pack.

ComfyUI Manager: search Runware, install, restart.

Manual:

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

No models to download (runware-sdk, pillow, soundfile). API key from runware.ai/api-keys, set in Settings or as RUNWARE_API_KEY.

Where people get burned

The trap here is over-tuning. With cfgInterval and modular scheduler exposed, it's tempting to treat music generation like a K-sampler and burn $10 chasing a perfect track. Start with the defaults, change one thing at a time, and treat advanced_json as a last resort - malformed JSON gets merged into the request and can fail the whole run. And the same ground rule as the base node: structured lyrics beat prose every time.