Paste your script, pick a voice, download the narration
0/20,000 characters
A script starts as text and has to end as narration. This page is that step: paste the script, pick a voice, and get back a spoken track you can drop onto a timeline. Nothing to install, no account to make, and no watermark read over the top of it.
The audio is generated on KlipTools' own servers with Supertonic 3, a compact open neural speech model of roughly 99 million parameters running through ONNX Runtime. The work does not happen on your machine, so a long script is no slower from a phone on a train than from a desktop.
One run takes up to 20,000 characters -- about twenty minutes of speech, which is a whole long-form script rather than a paragraph at a time. The result comes back as a 128 kbps MP3 and an uncompressed WAV from the same generation, so you can publish one and edit from the other.
Up to 20,000 characters per run, which is roughly twenty minutes of speech. For most channels that is a complete long-form script in a single job. If yours is longer, split it at a section break, generate each part, and merge them with the Audio Joiner -- a break between sections is where a pause belongs anyway.
Because the speech is generated on KlipTools' servers rather than in your browser, and those servers work through submitted jobs in order. The page shows your position in the queue and a progress bar while it runs, and the audio appears on the page when the job finishes. A short script is usually a short wait; a 20,000-character one takes longer for the same reason a longer video takes longer to export.
High runs more diffusion steps over the same text, which takes longer and produces a cleaner, steadier read. Standard is the faster of the two and is the sensible setting while you are still auditioning voices or building a scratch track to cut against. Switch to High for the version that actually goes in the published video.
The WAV. It is the uncompressed master from that generation, so anything you do to it afterwards -- levels, joins, a music bed under it -- starts from the full-quality file. The MP3 is 128 kbps and exists for the times you just need the narration somewhere quickly. Both come from the same generation, so they are the same performance.
Yes, and that is the main reason the voices are numbered. Pick a voice, write the number down, and use it again next episode: F3 is the same F3 next month, on either KlipTools site, with no profile or login tying it to you. Nothing about the voice drifts between sessions.
It is free every day, with no account, no signup, no card and no watermark. Every visitor gets a daily allowance of free credits shared across every KlipTools tool on both sites, and a voiceover costs 3 credits per character. When the allowance runs out you can come back tomorrow, or paste a KlipTools credits code to carry on immediately.
Six hours, then they are deleted. That is long enough to finish the edit you are in the middle of, and short enough that your narration is not sitting on a server for weeks. Download the MP3 and the WAV while you have them -- regenerating the same script later is a new job and costs credits again.
Yes. Supertonic's model weights are released under the OpenRAIL-M licence, and the audio you generate can go in monetised videos, client work, ads and products you sell. The one thing the licence rules out is using it to deceive, to cause harm, or to impersonate a real person without their consent.
No. There is no voice cloning here, and that is a decision rather than a missing feature. You get ten fixed, pre-trained voices; there is nowhere to upload a sample, so there is nothing anyone can do with this page that a person would have to be warned about.
Yes. The dropdown offers 17 languages and the model itself covers 31, so the script can be in whichever of them you write in. Every one of the ten voices speaks every language, which is what makes a translated version of a video sound like the same channel rather than a different one.
With the speed control, which runs from 0.7x to 2.0x, and with punctuation. A comma is a short pause and a full stop is a longer one, so the script you would read aloud yourself is the script that comes back sounding right. Changing the speed means generating again, so settle the wording first and the pace after.
Yes, and it is worth doing. Credits are charged per character, so auditioning a paragraph in two or three voices costs a fraction of what the full script does. Use the opening lines of the actual video rather than a test sentence -- how a voice handles your own writing is the thing you are judging.
Supertonic 3, an open neural text-to-speech model of about 99 million parameters, running under ONNX Runtime on KlipTools' servers. It is small for what it does, which is why a twenty-minute script is a wait of minutes rather than an overnight render, and why the same job costs the same whether you submit it from a laptop or a phone.
Two options, and the first one is usually better. Cut words out of the script and generate again: narration that fits because it says less almost always sounds better than narration squeezed to fit. If the shot is fixed and the words are not negotiable, nudge the speed up -- 1.1x is hard to notice, and the control goes to 2.0x if you need it.