I started APEXFlow after running into a few limitations while using the original ACE-Step UI.

In its Python fallback workflow, ACE-Step 1.5 Turbo was hard-wired, which meant there was no real way to switch between models. LRC not support , and the lyric experience was fairly basic. Generation progress was shown as a simple progress bar with no indication of what the model was actually doing, and the Spotify-inspired profile page felt more like a placeholder than a finished part of the product.

I originally planned to fix those specific issues, but the project gradually grew into a broader rebuild. I added model selection, clearer generation stages and status messages, synced lyrics, a more complete artist profile, and several new features around playback and local music management.

APEXFlow is still based on the original open-source project, but it now reflects the way I actually want to generate, review, and organize music locally.

I’ve just published the full walkthrough and source code below. 🎧