Used Suno to generate 500 background tracks for a stock audio site over the last quarter. Vendor marketing suggests it's a reliable, high-quality production tool.
My data says otherwise.
* **Consistency failure:** Claimed "studio-quality" output is not reproducible. Roughly 30% of generations had audible artifacts (muddy bass, metallic highs) making them unusable for sale.
* **SLA/throughput misalignment:** Promised "fast generation" doesn't account for queue times during peak hours. Actual time to usable batch was 2-3x longer than estimated.
* **Pricing opacity:** "Simple credit system" obscures true cost. Generating to a specific mood/brief often required 3-4 regenerations per usable track, blowing through credits.
* **License ambiguity:** Their terms state you own the output. But do they train on your prompts/generations? No clear public answer. Unacceptable for commercial resale.
Anyone else using it at volume for commercial purposes should verify:
* Actual usable output rate versus credits spent.
* How they handle and document training data exclusion.
* True generation time including server-side delays.
The license point is the real killer for stock audio. If they're training on your outputs, you're potentially devaluing your own catalogue by feeding their model. Their terms are silent because it benefits them.
You've hit the core issue with these platforms: the unit economics only work if their "average case" quality matches your commercial needs. A 30% discard rate on raw output, multiplied by 4x regens to hit spec, means your actual cost per usable track is 4-5x the advertised rate. That's not a tool, it's a slot machine.
Did you track the variance in discard rate by genre or tempo? I'd bet the artifacts are worse in complex arrangements.
The real cost is in the fine print.