Short answer: to add sound effects to a video with AI, upload the clip to Video Tools, then Generate Audio in Kolbo.AI and pick Mirelo SFX 2.0. It watches the footage and builds a full sound design synced to the action, keeps the dialogue, and handles clips up to 10 minutes with up to 4 variations. For short one-off effects and seamless loops up to 30 seconds, use ElevenLabs SFX v3 in Text to Sound.
Most AI video models now ship with some built-in audio, but plenty of footage still arrives silent: screen recordings, stock clips, older generations, phone footage with useless room noise, or a render where the model's own sound simply does not fit. Sound design is what makes a clip feel finished. This guide covers the two newest sound effects models in Kolbo.AI, what each one is built for, and the exact steps to use them.
The two models at a glance
Kolbo.AI added Mirelo SFX 2.0 in October 2026 for both Text to Sound and Video Tools, and ElevenLabs SFX v3 two weeks later as its own model in Text to Sound. They solve different problems, so the right pick depends on whether you start from a video or from a description.
| Mirelo SFX 2.0 | ElevenLabs SFX v3 | |
|---|---|---|
| Where in Kolbo.AI | Video Tools, Generate Audio (video in) and Text to Sound (text in) | Text to Sound |
| Input | A video clip, or a text prompt | A text prompt |
| Output | The same video with a new soundtrack, or an audio file | An audio file (mp3) |
| Maximum length | 10 minutes (600 seconds) | 30 seconds |
| Minimum length | 1 second (video), 0.1 seconds (text) | 0.5 seconds |
| Variations per run | 1 to 4 | 1 |
| Keeps existing dialogue | Yes, on by default (Keep Dialogue) | Not applicable |
| Seamless loop | No | Yes |
| Extra controls | Avoid (negative prompt), Seed, Length, output format and bitrate for text | Prompt Influence, auto or custom duration |
| Text output formats | mp3, wav, aac, flac | mp3 |
| Prompt length | Up to 2,000 characters (optional for video) | Up to 5,000 characters |
| Billing basis | Per second of output, per variation | Per second of delivered audio |
Live credit rates for every model are listed on the pricing facts page. At launch, Mirelo SFX 2.0 was set at 3.2 credits per second per variation and ElevenLabs SFX v3 at 0.43 credits per second, so a 30 second ElevenLabs effect comes to about 13 credits.
Mirelo SFX 2.0: sound design that watches your video
Mirelo SFX 2.0 is the model to reach for when you already have footage. In Video Tools, Generate Audio, it watches the clip and builds a layered sound design synced to the action on screen, built around any dialogue that is already there.
What makes it useful for real projects:
- Long clips. It accepts video up to 10 minutes and keeps the sound design consistent across the whole clip, instead of treating every few seconds as a separate job.
- Dialogue stays. Keep Dialogue is on by default, so the speech already in your clip is preserved and the effects are built around it. Turn it off when you want a completely new soundtrack.
- Up to 4 variations. Sound design is a matter of taste. One run can return up to four takes, so you can compare them and keep the one that fits.
- A prompt is optional. With no prompt, the model decides from the picture alone. Add a short direction ("heavy rain, distant thunder, no music") when you want to steer it.
- Avoid. The negative prompt field lets you exclude what you do not want, for example "music, speech".
- Length. By default the whole clip is scored. Set a shorter Length if you only need the opening seconds.
- Seed. Reuse a seed to get a repeatable result while you refine the prompt.
The same model also runs in Text to Sound. There it generates sound effects and full ambiences from a description, up to 10 minutes long, with the same variations, Avoid and Seed controls, plus a choice of mp3, wav, aac or flac and a bitrate for mp3 and aac.
ElevenLabs SFX v3: precise short effects and loops
ElevenLabs SFX v3 is the newest ElevenLabs sound effects model. It lives in Text to Sound next to the earlier ElevenLabs sound effects model, which is still available for fast everyday effects.
Where v3 stands out:
- Prompts with several sounds. It follows a prompt that mixes sounds more closely, such as "bright wind chimes tinkling in a breeze, then one deep gong strike".
- Music and voices inside an effect. It handles more convincing music and vocal sounds, for example a guitar riff with a person humming and a small crowd clapping.
- Cleaner high sounds. Chimes, gongs and whistles come out cleaner.
- Seamless loop. Switch on Seamless loop to get a sound that repeats without a seam, ideal for backgrounds, ambience beds and game audio.
- Duration from 0.5 to 30 seconds, or leave it on auto and let the model choose.
- Prompt Influence controls how strictly the model sticks to your words.
Because it tops out at 30 seconds and returns an audio file rather than a scored video, v3 is a sound library builder: whooshes, impacts, UI clicks, transitions and loops that you then place on a timeline.
How to add sound effects to a video in Kolbo.AI
- Open Video Tools. Sign in at app.kolbo.ai, open Video Tools and choose Generate Audio.
- Upload your clip. Any video from 1 second up to 10 minutes.
- Select Mirelo SFX 2.0 as the model.
- Decide on dialogue. Leave Keep Dialogue on if the clip has speech you want to keep. Turn it off for a full replacement soundtrack.
- Write an optional prompt. Describe the mood or key sounds in plain words. Put anything you do not want, such as music, in Avoid.
- Pick the number of Variations. One is enough for a quick pass. Two to four gives you options for a client or a final cut. The cost card multiplies by the number of variations before you generate.
- Optionally set Length and Seed. Use Length to score only part of the clip and Seed to reproduce a result.
- Generate and compare. Play each variation and download the one that fits.
How to create a single sound effect or loop
- Open Audio Tools, then Text to Sound.
- Choose ElevenLabs SFX v3 for short, precise effects and loops, or Mirelo SFX 2.0 for longer ambiences and layered sound design.
- Describe the sound in order of events: what happens first, then what follows.
- With ElevenLabs SFX v3, set a duration (0.5 to 30 seconds) or leave it on auto, and switch on Seamless loop for a background bed.
- With Mirelo SFX 2.0, set the duration (up to 10 minutes), the number of variations and the output format.
- Generate, listen and drop the file into your edit.
Which one should you use?
| Your job | Best pick |
|---|---|
| A silent or badly recorded clip that needs a full soundtrack | Mirelo SFX 2.0 in Video Tools |
| A talking-head or interview clip that needs atmosphere under the voice | Mirelo SFX 2.0 with Keep Dialogue on |
| A long video (minutes, not seconds) | Mirelo SFX 2.0, up to 10 minutes |
| Several sound design options to choose from | Mirelo SFX 2.0, up to 4 variations |
| A one-shot whoosh, hit or UI sound | ElevenLabs SFX v3 |
| A looping ambience for a game, app or background | ElevenLabs SFX v3 with Seamless loop |
| A long ambient track from a description | Mirelo SFX 2.0 in Text to Sound |
A common workflow combines both: score the whole video with Mirelo SFX 2.0, then add a few precise accents from ElevenLabs SFX v3 on top in your editor.
Prompt tips for better sound effects
- Describe sources, not feelings. "Gravel crunching under boots, wind through pine trees" beats "tense outdoor vibe".
- Order the events. Write them in the order they happen: "a door creaks open, then a cat meows".
- Keep it to two main events per short effect. When we prepared the ElevenLabs SFX v3 launch samples, two-event prompts came out reliably, while three-event prompts sometimes dropped one of the sounds.
- Use Avoid for what keeps sneaking in. If music shows up where you wanted pure ambience, add "music" to Avoid in Mirelo SFX 2.0.
- Generate variations instead of rewriting. With Mirelo SFX 2.0, two to four variations of a good prompt often beats five rewrites of a single take.
Where this fits in the rest of Kolbo.AI audio
Sound effects are one piece of the audio stack. For music, voices and transcription, see the complete guide to Kolbo.AI audio tools. If you would rather pick from ready-made sounds, the built-in AI sound effects and music library lets you search by feel. Some video models also generate sound in the same pass, covered in Seedance 1.5 with built-in sound. When the clip needs new speech rather than new effects, the AI lipsync and dubbing guide is the next step.
Mirelo SFX 2.0 is also available through the Kolbo.AI SDK and MCP server with the same settings, so an agent or script can score videos the same way the app does.
FAQ
How do I add sound effects to a video with AI?
Open Video Tools in Kolbo.AI, choose Generate Audio, upload your clip and select Mirelo SFX 2.0. The model watches the video and returns it with a synced sound design. Keep Dialogue preserves existing speech, and you can request up to 4 variations.
What is the best AI sound effects generator for long videos?
In Kolbo.AI, Mirelo SFX 2.0 handles clips up to 10 minutes and keeps the sound design consistent across the whole clip. ElevenLabs SFX v3 is limited to 30 seconds per effect, so it suits short effects and loops rather than full videos.
Will AI sound effects remove the dialogue in my video?
Not with Keep Dialogue on, which is the default for Mirelo SFX 2.0 in Video Tools. The speech stays and the effects are built around it. Switch it off only when you want a fully new soundtrack.
Can I make a seamless looping sound effect with AI?
Yes. In Text to Sound, choose ElevenLabs SFX v3 and turn on Seamless loop. It produces a sound that repeats without an audible seam, up to 30 seconds long, which works well for ambience beds, menus and game backgrounds.
What is the difference between Mirelo SFX 2.0 and ElevenLabs SFX v3?
Mirelo SFX 2.0 can watch a video and score it, runs up to 10 minutes and returns up to 4 variations. ElevenLabs SFX v3 works from text only, runs up to 30 seconds, adds seamless loops and Prompt Influence, and is strong on prompts that mix several sounds.
How much does it cost to add AI sound effects to a video?
Both models are billed per second of audio produced. Mirelo SFX 2.0 also multiplies by the number of variations, and the cost card shows the total before you generate. Current rates are on the pricing facts page.
Can I use AI sound effects without a video?
Yes. Text to Sound generates audio files from a description with either model: ElevenLabs SFX v3 for short effects and loops, Mirelo SFX 2.0 for longer ambiences up to 10 minutes in mp3, wav, aac or flac.