ElevenLabs: your first 30–60 second voiceover (2026)

ElevenLabs does not write your message — it reads it aloud. This flow takes 15 minutes and produces a 30–60 second voiceover you can drop on a Runway clip or a pitch slide. You start from a short script that already works — ChatGPT or by hand — and pick a library voice, not a clone.

1. Steps 1

   Write a 4–6 sentence script (30–60 s)

   Open ChatGPT or a text editor and write 4–6 short sentences. One idea per sentence. Aim for 30–60 seconds of voiceover: roughly 80–120 words in English, a bit less in Spanish. Write numbers and symbols as words («three out of five», not «3/5»). Audio language comes from the text; accent comes from the voice you pick later. If the script is for a product clip, state benefit and call to action, not empty adjectives.

   What you should see: A block of 4–6 sentences in an editor or ChatGPT, ready to copy. ElevenLabs not open yet.

2. Steps 2

   Open ElevenLabs → Text to Speech

   Go to elevenlabs.io and sign in. In the ElevenCreative sidebar, choose Text to Speech (or go straight to elevenlabs.io/app/speech-synthesis/text-to-speech). Paste your script in the central text field. Do not open Voice Cloning, Studio or the API — this tutorial stays on basic web synthesis.

   What you should see: The Text to Speech editor with your script pasted, voice field empty or on a default voice, and empty history on the right.

3. Steps 3

   Pick a library voice that matches the tone

   Bottom left, open the voice selector (Voices). Browse Default Voices or Voice Library. Play previews: look for conversational, narrator or formal tone to match your script. For Spanish voiceover, pick a Spanish-native library voice — accent comes from the voice, not a separate language toggle. Voice matters more than model; if none fit, try another before touching sliders.

   What you should see: A voice selected with its name visible in the selector. Playable preview before you generate.

4. Steps 4

   Model and starting sliders

   In the model selector, pick Eleven v3 for a short expressive take (supports tags like [whispers] or [sighs], up to 5,000 characters on paid plans). For something steadier across sentences, Multilingual v2 also works. Do not start with Flash v2.5: it is cheap and fast, not the best first pick for a voiceover. Set sliders per docs: Stability ~50, Similarity ~75, Style 0. Generate when script, voice and model are set.

   What you should see: Eleven v3 or Multilingual v2 selected. Sliders at starting values. Generate button ready.

5. Steps 5

   Generate 2–3 takes and pick the best

   Hit Generate and wait for audio. Play the full take. If the overall pace works but one sentence sounds off, do not regenerate the same text yet — note which line fails. For a second take, tweak the script (an extra comma, a shorter sentence) or run the same setup once more to compare model variation. Keep the take where the message is clear start to finish.

   What you should see: One or two waveforms in the history on the right. Audio playable from start to finish.

6. Steps 6

   Rewrite the bad line, do not regenerate identical text

   If a sentence sounds flat or rushed, edit the script: add a comma, split into two short sentences, change a verb or write numbers as words. Example: «Our coffee reaches your door in forty-eight hours» instead of «48h delivery». Hit Generate again with the edited text. Regenerating the same paragraph with the same voice burns credits with no guarantee of improvement — intonation changes more with rewriting than with clicking again.

   What you should see: Edited script in the text field with at least one reformulated sentence. New generation in history.

7. Steps 7

   Download MP3 and listen over the real clip

   When you have the keeper take, hit Download by the player or go to History → pick the entry → MP3 128 kbps. For WAV, M4A or FLAC, use Advanced in history. Name the file by project (`pitch-coffee-v2.mp3`). Play it over your Runway clip or slide in whatever editor you use (CapCut, Premiere, Keynote). Syncing audio and video is not ElevenLabs — that is the editor’s job.

   What you should see: An MP3 in your downloads folder. Audio roughly synced over video or slide in a player or timeline.