Skip to content
Notifications
Clear all

My results after generating 100 tracks for a meditation app - success rate?

1 Posts
1 Users
0 Reactions
2 Views
(@code_reviewer_anna)
Estimable Member
Joined: 3 months ago
Posts: 122
Topic starter   [#11040]

Hey folks! 👋 I've been using Suno for a few months now, mostly to generate background music for a meditation app side project. I wanted to get some hard data on its practical reliability, so I decided to run a little experiment: generating 100 tracks with specific parameters and logging the results.

Here's the basic prompt structure I used, aiming for consistent, non-vocal, ambient pieces:

```python
prompt_template = {
"genre": "ambient, atmospheric, calm",
"mood": "peaceful, serene, drifting",
"instruments": "piano, soft pads, gentle strings",
"key": "mostly minor or neutral",
"duration": "2-3 minutes",
"vocal_style": "no vocals"
}
```

**Of the 100 generations:**
* **72** were usable right out of the box. Perfect length, no weird tempo shifts or sudden dramatic elements.
* **18** had minor issues – a slightly jarring transition, or an instrument that didn't quite fit. These were often salvageable with a re-roll on a specific section.
* **7** were complete misses. Think random upbeat electronic beats or, in two confusing cases, faint spoken word that wasn't in the prompt.
* **3** generations failed technically (the generation errored out).

**My main takeaways on workflow:**
* Being **hyper-specific** about what you *don't* want (e.g., "no drums, no vocals, no sudden changes") improved consistency dramatically.
* The "Custom Mode" with advanced settings was essential for locking down the style after I found a good seed.
* The re-roll feature for individual song sections is a lifesaver. It turned many of those "minor issue" tracks into usable ones without starting from scratch.

The ~75% direct success rate feels pretty solid for a tool like this, though it requires a curated prompt and a willingness to do some light "gardening" of the outputs. For my use case, it's a huge time-saver compared to sourcing/licensing library music.

Has anyone else done a similar volume test? I'm curious if your success rates align, especially in other genres. Also, any tips for making those prompts even more bulletproof?


Clean code is not an option, it's a sanity measure.


   
Quote