ACEStep 1.0 Repaint Guider BETA ⚡🅡🅞🅣🅘

Read the display name carefully: ACEStep 1.0 Repaint Guider BETA. This targets ACE-Step 1.0 latents specifically, and it's flagged BETA by its own author. It's worth knowing upfront because this pack's active development has moved on to native ACE-Step 1.5 guiders (Edit/Cover) - this node is the older, legacy path, kept around because plenty of people still have ACE-Step 1.0 workflows and latents they want to touch up. If you're starting a new music-generation workflow from scratch on ACE-Step 1.5, look at the pack's 1.5 Edit guider first; this one is for when you specifically need to work with 1.0.

What it does

You generated a track (or have source_latents from one), and you want to regenerate just a window of it - fix a bad section, redo the bridge - without touching everything outside that window. This node builds a GUIDER that does exactly that: selectively repaint [start_time, end_time] while preserving the rest of the latent, with a soft blend at the boundary so the transition doesn't click.

Unlike most nodes in this pack, this one sits in ComfyUI's own native sampling category - it plugs directly into SamplerCustomAdvanced the same way a built-in CFGGuider would, alongside your sampler and scheduler nodes.

Inputs and outputs that matter

• model, positive, negative - your loaded ACE-Step 1.0 model and conditioning, same as any sampling setup.
• source_latents - the already-generated audio latent you're repainting into.
• start_time / end_time - the region to repaint, in seconds (0-1000, defaults 10/20).
• repaint_strength (0.0-1.0, default 0.7) - how much of that window gets overwritten versus preserved; higher values lean harder toward the new generation inside the window.
• feather_time (0.0-2.0, default 0.1) - softens the boundary of the repaint window so the edit blends instead of cutting cleanly, avoiding an audible seam.
• cfg (0-100, default 1) - standard guidance scale for the repaint pass.

Output is a single GUIDER, which you wire into SamplerCustomAdvanced like any other guider.

Installing it

ComfyUI Manager: search RyanOnTheInside, install, restart. Manual:

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

This node needs an ACE-Step 1.0 checkpoint loaded separately - the pack doesn't ship the model itself, just the guider that drives it.

Where people get burned

start_time/end_time accept values up to 1000 seconds, but your actual source_latents probably don't cover anywhere near that - check the real duration first (AudioInfo on the source audio, before it was encoded to a latent) or you'll target time that doesn't exist in your latent. And treat the BETA label as accurate, not boilerplate: this is a legacy-model guider the author kept working rather than the current focus of development, so if you hit a rough edge, check whether the same thing you're trying to do is better served by the pack's newer ACE-Step 1.5 Edit/Cover guiders before assuming this one is the only path.