AI video generation has been moving at light speed. The visuals are stunning — but there is something missing.

Sound.

Most AI video providers generate videos without sound. No matter how cinematic, a visual without sound feels unfinished. That changes today. Meet Mirelo SFX, the first sound effects model designed specifically for AI-generated video.

Available now on fal.ai.

Sound generated by Mirelo SFX.

The Model That Unmutes AI Videos

Before, adding sounds to AI videos was a tedious process. Now, Mirelo SFX lets you easily create perfectly synced cinematic sound with one click.

• The video is the prompt — Just pass in a video. No text descriptions, no extra setup. Mirelo SFX generates perfectly timed sound directly from the visuals.

• High sound quality — We focus on production-ready cinematic sound.

• Pure effects, no noise — Forget the leakage of unwanted background music or speech. You get clean, contextual sound effects only.

• Optimized for synthetic video content — Mirelo SFX thrives where others stumble.

Typically, AI-generated videos don’t come with sound. Mirelo SFX unmutes them with one click.

Built for production pipelines

By integrating Mirelo SFX through fal.ai, video-gen platforms can now deliver complete experiences by default:

Real-time generation — 10 seconds of video = 10 seconds of compute.
Multiple variations — 2–8 outputs per clip so users can pick their favorite.

Mirelo SFX is the new world’s best video-to-sfx model

The uncanny valley is finally crossed. In listening tests, users preferred Mirelo SFX’s output 67-77% of the time. This isn’t a nice-to-have anymore—it’s a model that can be chained by default, knowing it will enhance every video, not degrade it.

Comparison study

In blind pairwise comparison tests, independent raters* compared generations from Mirelo SFX and three of the most popular existing video-to-sfx alternatives (PixVerse, MMAudio V2 and ThinkSound). Mirelo SFX achieved a total win rate of 67.4% (excluding ties) and 77% (including ties), clearly outperforming all three alternatives.

Side-by-side examples

Competitors often struggle with unwanted artifacts like music or speech, particularly on synthetic videos (see, e.g., examples 1 and 2 below). Mirelo SFX also excels at creating realistic effects across a multitude of scenes when alternative models default to unrelated or unnatural sounds (see examples 2 and 3 below).

Try it yourself!

👉 Try it live in the fal.ai playground

👉 Try it within the Mirelo Studio

What’s next?

We are constantly pushing updates to our models and ship new products — so stay tuned. Follow us on social media to never miss new updates: Linkedin, X, Instagram, TikTok.

*We recruited 20 external listeners from a crowd-sourcing platform via mabyduck. The raters were pre-screened by mabyduck to ensure they listen closely to the given audio. Raters performed pairwise comparisons between different audio generations for a given video, with the option to vote for a “tie”. 740 pairwise comparisons were collected in total.