Chitra Studio
Sound effects17 October 2026
One sound effect, five AI models
Only two of five did what I asked. Comparing Stable Audio 3 SFX, ElevenLabs Sound Effects v2, CassetteAI, Mirelo SFX 2.0 and MMAudio v2.
Sound effects are the cheapest part of AI media, and the easiest to get wrong. A bark that sounds like a cough, or a street that sounds like a hairdryer, gives away a clip that looks perfect.
I gave five sound-effect models the same prompt. Here they are side by side, best first, with what each cost and what I heard.
Making a sound effect
- Switch the prompt bar to audio and choose Sound Effect.
- Describe the sound: what it is, how close, and what is behind it.
- Choose the model and the Length, then press Generate.
The prompt
A dog barks sharply four times in ten seconds, loud and close in the foreground. Quieter ambient traffic noise, distant car engines, and hum of a busy city street at night.
Five models, one prompt
Stable Audio 3 SFX
5 s long · 3 credits · made in 9 s
The best of the five: exactly four barks, as asked, close and in front of the traffic.
Built for distinct events such as barks or knocks, up to two minutes.
ElevenLabs Sound Effects v2
5 s long · 2 credits · made in 6 s
Close behind: clear, sharp barks, but eight of them if you count each double bark, not four.
Built for crisp single effects, and the only one here that can loop, up to 22 seconds.
CassetteAI SFX
5 s long · 2 credits · made in 8 s
One bark, and the rest is mostly noise.
Built for quick textures for drafts, up to 30 seconds.
Mirelo SFX 2.0
10 s long · 14 credits · made in 8 s
One bark in ten seconds, and mostly silence. The dearest of the five, and nearly the least useful.
Built for steady scenes and long ambiences rather than single events, up to two minutes.
MMAudio v2
8 s long · 2 credits · made in 36 s
Pure noise. No bark at all.
Built for steady textures and ambiences at the lowest price, up to 30 seconds.
All five together cost 23 credits ($0.23). The "built for" notes come from Chitra Studio's model list.
What I took from it
- Only two of five did what I asked. Stable Audio 3 SFX and ElevenLabs Sound Effects v2 made a dog barking in traffic. The other three gave me one bark, or none, in a wash of noise or silence.
- Pick a model built for the job. A bark is a distinct event. The model built for distinct events got the count exactly right; the two built for steady ambiences barely barked at all.
- Numbers in a prompt are a request, not a rule. I asked for four barks. Stable Audio gave four, ElevenLabs gave eight.
- Price does not buy accuracy. The best result cost 3 credits; Mirelo SFX 2.0 cost 14 and was one of the worst.
- The length is a setting, not a word in the prompt. I wrote "ten seconds", but three of these are five seconds long, because that is the length I had set.
Tips
- Run the same prompt on two or three models. At 2 or 3 credits each, hearing it three ways costs less than one wrong pick.
- Listen before you use it. A sound can look fine as a waveform and still be the wrong thing.
- Say how close. "Loud and close in the foreground" and "quieter in the background" tell the model how to place each sound.
- Set the length to match the clip you are putting it under.
Make sound effects for your own clips
Chitra Studio is a Mac app for making and organising AI images, video and sound with more than 60 models. Every price is shown before you generate.
Download for Mac14-day free trial · 50 free credits · no card