Generate Your First Music Track

In this tutorial you’ll install ACE-Step, configure a few settings, and generate a 30-second AI music track — all in about 10 minutes.

~10 minBeginnerYou'll build: A 30-second AI-generated music track

Step 1: Prerequisites

ACE-Step requires a GPU with at least 8 GB VRAM. AMD APU users (e.g. Ryzen 780M): the first run takes several extra minutes while MIOpen builds its find-db cache — subsequent runs are much faster.
CPU-only mode is not supported for ACE-Step. A discrete or integrated GPU is required for music generation.

Make sure you have:

  • Thumper-Run installed (installation guide)
  • At least 10 GB free disk space (app + models)
  • A GPU with 8+ GB VRAM (or AMD APU with shared memory)

Step 2: Install ACE-Step

ACE-Step is a fast, high-quality AI music generator. Let’s get it installed.

  1. In the catalog, find ACE-Step (or type "ace" in the search bar)
  2. Click the Install button on the app card
  3. Wait for models to download — approximately 5 GB total
console
Downloading ACE-Step models...
  ace_step_v1.5_384m.pth     2.1 GB  [================] 100%
  vocoder.pth                 0.3 GB  [================] 100%
  text_encoder/               2.6 GB  [================] 100%
Installation complete.

Test it: After installation finishes, confirm the ACE-Step card shows a green "Installed" badge in the catalog.

If you see an out-of-memory (OOM) error during installation, close other GPU-intensive applications and try again. The model files need to be fully downloaded before first launch.

Step 3: Generate Your First Track

Time to make some music!

  1. Click Launch on the ACE-Step card
  2. Set duration to 10 seconds (for a quick first test)
  3. Choose Pop as the genre
  4. Click Generate and wait for the track to complete

Generation times vary significantly by GPU. Here’s what to expect:

GPU10s Track30s Track60s Track
RTX 4090~2s~5s~9s
RX 7900 XT~4s~10s~18s
AMD 780M (APU)~43s~90s~160s

Test it: Play back the generated audio directly in the ACE-Step UI. You should hear a short pop melody.

AMD GPU users: if you hear silence or the output contains NaN values, switch to BF16 precision. Navigate to the app detail page → Settings tab → Precision dropdown → select BF16. FP16 causes overflow in the DiT attention layers and must not be used on AMD hardware.

Quality Check

After generation, listen to the full track before moving on. Common quality issues to watch for: silence gaps (usually a precision issue), repetitive loops (try a longer duration or different genre), and clipping artifacts (reduce the volume in post-processing). If the output is completely silent, check the console for NaN warnings — this almost always means FP16 is selected on AMD hardware.

Output Specs

Generated tracks are saved as uncompressed WAV files: 44.1 kHz sample rate, 16-bit depth, stereo. This is CD quality and suitable for most production workflows. Convert to MP3 or FLAC externally if needed.

If generation appears to hang for more than 5 minutes on a supported GPU, the process may have stalled. Stop the app from the Thumper-Run dashboard and relaunch. The first run on AMD APUs can legitimately take several minutes due to MIOpen cache warmup — open the Thumper-Run dashboard sidebar → Console to check for MIOpen "find-db" progress lines before assuming a hang. Once the cache is built, subsequent runs skip this step entirely.

Step 4: Customize Your Track

Now let’s explore the controls to shape your music.

Duration

Set the track length in seconds. Start short (10–30s) while experimenting, then go longer once you’re happy with the style.

Genre

ACE-Step supports a wide range of genres. Here are some popular starting points:

GenreBPM RangeDurationBest For
Pop100–13030–60sCatchy melodies, vocals
Electronic128–15030–60sInstrumentals, loops
Jazz80–12030–60sSmooth backgrounds
Classical60–10030–120sOrchestral pieces
Hip-Hop80–11030–60sBeats, rap vocals

The genre tag guides the model’s style, instrumentation, and rhythm.

Lyrics

Add lyrics in the text box to generate vocal tracks. Leave it empty for instrumental-only output. Lyrics are time-aligned automatically — no manual syncing needed.

BPM

Beats per minute controls the tempo. Typical ranges: 60–80 for ballads, 100–130 for pop, 130–160 for electronic/dance.

Tracks longer than 60 seconds may cause out-of-memory errors on GPUs with less than 16 GB VRAM. If you hit OOM, reduce the duration or close other GPU applications.

Step 5: Export Your Track

Save your creation and find it later.

  1. Click the Download button below the audio player
  2. The track saves as a WAV file to your configured output directory
  3. Open the Gallery page from the sidebar to browse all generated tracks

Default output location:

~/.local/share/tr-desktop/outputs/ace-step/

Key Takeaways

  • ACE-Step requires a GPU with 8+ GB VRAM — CPU-only is not supported
  • AMD APU first-run is slow due to MIOpen cache warmup; subsequent runs are much faster
  • Always use BF16 precision on AMD GPUs to avoid NaN/silence issues
  • Keep tracks under 60 seconds on GPUs with less than 16 GB VRAM to avoid OOM
  • Generated tracks are saved as WAV files and indexed in the Gallery automatically

Next Steps

Learn More

First Image Tutorial

Generate AI images with ComfyUI

First Chatbot Tutorial

Chat with a local LLM

User Guide

Full app management documentation

Troubleshooting

GPU issues and benchmarks