Candy Dungeon Music Forge

Candy Dungeon Music Forge (CDMF) is a local AI music workstation for people who actually like owning their tools.

CDMF runs on ACE-Step under the hood, but it’s more than “just a wrapper” or another generic node graph. It’s a focused, sleek UX for actually using ACE-Step day to day: generating tracks, organizing them, and training custom LoRAs without getting lost in boilerplate.

Write a prompt → get a full track. Train LoRAs on your own datasets. Keep everything on your own hardware.

Official links:
Official CDMF website (musicforge.candydungeon.com)
Candy Dungeon project hub (candydungeon.com)

🧁 What CDMF does

• Text-to-music engine (ACE-Step based)
  Generate fantasy, chiptune, lo-fi, ambience, and more from plain-English prompts. CDMF ships with sane defaults tuned for high quality within ACE-Step’s limits, so you can get great results quickly without hand-tuning a dozen sliders.
• Curated library & media player
  Every track you generate is stored in a built-in browser where you can favorite, tag, filter, and categorize your songs. Copy generation settings and presets between tracks with a click so you can iterate on a vibe or share recipes with friends.
• LoRA training UI & dataset tools
  Train your own LoRA adapters on curated datasets to create signature “sound packs” and project-specific vibes. CDMF includes tools for mass tagging and organizing datasets, so preparing training data is part of the workflow instead of an afterthought.
• Core + advanced controls with tooltips
  The generation panel is split into “core” settings (for fast results) and “advanced” controls (for power users). Inline tooltips explain what each knob does, so you can grow from casual user to expert without needing to read the ACE-Step source code.
• Stem control
  Built-in stem separation lets you rebalance vocals vs. instrumentals or export instrumental-only mixes directly from the UI.
• Local-first by design
  No cloud GPU credits, no uploads. CDMF embeds Python and sets up a virtual environment on first launch, then keeps everything running on your machine.

💾 System requirements

Minimum

• Windows 10/11 (64-bit)
• NVIDIA GPU (RTX strongly recommended)
• ~10–12 GB VRAM (more is better)
• SSD with tens of GB free

Comfortable experience

• Modern RTX card with plenty of VRAM
• 32 GB system RAM
• SSD for models & datasets
• Comfortable using “power user” tools

First-time setup can take a while while CDMF installs Python packages and downloads models. After that, launches are much faster.

💸 Pricing & early access

CDMF is currently in early access.

• Launch price: $19 USD
• Planned regular price: $29 USD

Early buyers get:

• The current CDMF installer
• Access to future bug-fix updates
• Example presets & LoRA configs
• New LoRAs trained on high-quality per-genre datasets (as they’re added)

If you’re the kind of person who likes playing with weird, powerful tools that run on your own GPU, you’re exactly who this is for.