A picture is half a film.
Now it makes the other half.
The Audio module is included in the $49 app: 272 voices you can audition free, voice cloning for $1.50, music beds from 2¢ a minute. The Audio tab adds the three things a finished piece still needed: a voice to narrate it, music to carry it, and the sound effects that sell it, plus a scorer that writes to a cut you've already made. Four lanes, the same model picker, the same exact price on the button, and the same "it just saves to a folder" ending.
272 voices. Hear every one before you spend a cent.
Most tools make you pay to find out what a voice sounds like. Every voice here ships with a real recording of the same line, loudness-matched so none of them wins on volume alone. Play as many as you like, free, before you've even entered a key.
- Six engines in one list: MiniMax Speech 2.8 HD, ElevenLabs Multilingual v2, Inworld TTS 1.5, Gemini 3.1 Flash, Seed Speech 2 and xAI
- Filter by gender, pitch and model, or type a name and search
- Curated, not scraped. Cartoon rosters and low-quality open models were cut on purpose. Every voice on this list is one you'd put in a client's video
- From a fraction of a cent: a 150-word script runs under 1¢ on Inworld, about 9¢ on MiniMax HD, quoted on the button before you click
- Pauses on cue. Write
(pause), or(pause 1.5)for a timed beat, anywhere in the script
Clone your own voice. Keep it forever.
Read a minute of script into your phone and get a voice that's yours, saved to a library and waiting in every future project. Asset Lab includes the script to read and the recording technique that makes the difference: written to cover every English sound, three kinds of sentence and the spoken numbers clones usually fumble.
- $1.50 per clone, once. After that it narrates at the price of ordinary speech
- Or design one from a description ("warm, gravelly, late-night radio") for $3.00
- Your library, protected. A cloned voice is deleted at the vendor if it goes seven days unused. Asset Lab keeps yours alive automatically, warns you before one lapses, and backs the whole list up to a file
- Consent, up front. The app asks you to confirm you own the recording or have the speaker's permission, every time
Pauses where you want them. A voice for every character. New in 2.6
Timing is half of a read. Type (pause) for a natural beat or (pause 1.5) for an
exact one, anywhere in a line, in the Voice lane or in a Director cut.
- Exact silence on MiniMax voices (every cloned voice included) and ElevenLabs; a natural breath on engines without pause control
- A voice per character. In a storyboarded scene, planned lines are spoken by whoever is on screen and narrated when nobody is; unassigned lines fall back to the narrator
- Make a still talk: a speaking character's approved storyboard frame becomes a lip-synced shot with one button
- Narrate a Director cut. Pick one voice for the whole film, write a line on any shot, and each read lands at its scene in the final mix
- Mixed, not stacked. Director's ♪ Music and 👤 Voice sliders set the balance you hear in the preview and get in the export
Music you describe in chips, not paragraphs.
Pick a genre, a mood, the instruments, a tempo and how the arrangement should move, and the studio assembles the structured caption the model actually wants. A disclosure panel shows the exact text before it goes. No black box.
- Six music models: MiniMax Music 3, ElevenLabs Music, Lyria 3 and 3 Pro, Sonilo v1.1 and CassetteAI
- Beds from 2¢ a minute, up to five-minute tracks, instrumental or with vocals
- Honest lengths and honest prices. A model that treats your slider as a maximum says so and quotes "up to", and one that bills per whole minute tells you a 20-second sting costs a full one
- Licensed option: Sonilo bills itself as commercial-safe, and it's flagged in the picker for client work
Hand it a cut. Get music written to the edit.
The Score lane starts from footage instead of a description. Point it at a clip (one from your gallery, or any file on your drive) and it reads the pacing and the cuts and writes music that lands with them. Score with no prompt and Asset Lab reads the footage first (era, setting, energy), then writes a period-accurate brief.
- Straight from your gallery, since anything you've generated is already there, no re-upload
- Or upload your own. Footage you shot works exactly the same way
- Up to three takes: three different scores for the same clip, priced per take so the total is never a surprise
- About 27¢ for a 30-second cue, and "Score this clip" sits in the gallery lightbox too, so you can start from the shot you're looking at
Give the voice a face.
Foley the scene for a penny.
Any generated or cloned voiceover can be sent to the Video tab and given a face. Attach a still and a character delivers your words; attach a clip you already made and it re-voices the footage. The output always matches your audio's length, so the price on the button is exact, not estimated.
Still + voice
Flashtalk drafts at 2¢/s, sync-3 Avatar also handles illustrations, and OmniHuman 1.5 makes emotion and gesture follow the voice.
Clip + voice
HeyGen from 5¢/s with optional burnt-in captions, HeyGen Precision, or sync-3 when the voice and the picture are different lengths.
Sound effects for a penny
A five-second effect costs about 1¢, so roll four and keep the best. Seamless looping for ambience that has to run under a scene.
Written for the ear
Surprise, History, Clear and ✦ Enhance work here too, but Enhance rewrites a script for reading aloud, and a music idea into a real brief.
Lands in ~/Music
Audio saves to ~/Music/Asset Lab, where a sound file belongs, while stills and clips stay in Pictures. Projects still group the lot.
Waveforms in the gallery
Every take gets a waveform tile you can spot at a glance and a player in the lightbox, with the same search, favourites and projects as everything else.
Commercial-use models
Every audio model is marked commercial-use on fal. Lyria 3 output carries Google's SynthID provenance watermark.
Drops onto the timeline
In AssetLab Resolve, a finished music bed or talking clip lands on your current timeline like any other render.
Narrate an explainer, bed a product film, foley a scene, or hand a finished cut to the scorer and let it write to your edit. Then give the voice a face and let a character deliver it. Included in both editions at no extra cost: the standalone studio and AssetLab Resolve.
Cents, quoted before you click.
Every rate below is fal.ai's raw price, billed to your own key with no markup. The exact figure for your script, length or clip sits on the Generate button.
| Lane | Models | Rate | Typical job |
|---|---|---|---|
| Voice | Inworld TTS 1.5 · xAI · Seed Speech 2 · Gemini 3.1 Flash · MiniMax Speech 2.8 HD · ElevenLabs Multilingual v2 | $0.01–$0.10 per 1,000 characters | 150-word script ≈ under 1¢ to 9¢ |
| Voice cloning | MiniMax voice clone · voice design | $1.50 per clone · $3.00 per designed voice | One-time; the clone then speaks at ordinary rates |
| Music | MiniMax Music 3 · ElevenLabs Music · Lyria 3 · Lyria 3 Pro · Lyria 3.5 · Sonilo v1.1 · CassetteAI | From $0.002/s · $0.04–$0.10 flat · $0.60/min | 60-second bed ≈ 2¢ to 60¢ by model |
| Sound FX | ElevenLabs SFX v2 · Stable Audio 2.5 | $0.002/s · $0.20 flat | Five-second effect ≈ 1¢ |
| Score | Sonilo v1.1 video-to-music | $0.009 per second per take | 30-second cue ≈ 27¢ |
| Lip sync | Flashtalk · sync-3 Avatar · OmniHuman 1.5 · HeyGen Lipsync · HeyGen Precision · sync-3 Lipsync | $0.02–$0.16 per second of audio | 10-second take ≈ 20¢ to $1.60 |
Rates re-verified against fal.ai each release. The full catalog, image and video included, is on the rate card.
About the Audio module
Can it make voiceover, music and sound effects?
Yes. The Audio tab has four lanes: Voice, Music, Sound FX and Score. 272 curated English voices across ten engines, and auditioning them is free. Every voice ships with a bundled recording, so you can play the whole list before spending a cent. Clone your own voice for $1.50, or design one from a description for $3.00. Music beds start around 2¢ a minute, sound effects about 1¢ for five seconds, and the Score lane writes music to an existing clip's pacing for roughly 27¢ per 30-second cue. Included in both editions at no extra cost.
Can it make a character say my voiceover, with lip sync?
Yes. Generate or clone a voice, then open the take and hit Make a talking video. Attach a face and a character delivers your words, lip-synced, in your voice. Flashtalk drafts at 2¢ a second, sync-3 Avatar also handles illustrations, and OmniHuman 1.5 makes emotion and gesture follow the voice. Drop in a clip instead of a face and it re-voices footage you already made, from 5¢ a second. The price is exact, not estimated, because the app measured your take when it drew the waveform.
Can I add pauses to a voiceover?
Yes. Write (pause) (or (pause 1.5) for a timed beat) anywhere in a line, in the Voice lane
or in a Director narration line. MiniMax voices (every cloned voice included) and ElevenLabs render exact silence;
engines without pause control take a natural breath instead.
Is previewing a voice free?
Yes. Every voice ships with a bundled, loudness-matched preview that plays with no API key and no charge. You only pay when you generate a take.
What happens to a cloned voice if I don't use it for a while?
MiniMax deletes a clone that goes seven days unused, and fal has no way to list your voices. Asset Lab renews your clones automatically, warns you from day five, and can back the whole list up to a file.
Are the reference video models lip-sync models?
No. Seedance 2.5 References and MiniMax H3 References accept audio, but they use it as a timing and beat signal for choreographing action and generate their own speech. Only the six lip-sync models above put your own recording in a character's mouth.
Voice, music, sound. Included.
The Audio module ships in both editions of Asset Lab: $49 once, no separate plan, no add-on.