How to add voiceover to an architecture video: a 30-second script, phone recording, AI voices and mixing

A project video without a voice relies on the images to explain themselves, and most of the time they can. But some projects have a story the pictures can't tell: a tight lot, a client's brief, a material chosen for a reason. A short voiceover can carry that, if it fits the length, sounds like a person in a quiet room, and sits on top of the music rather than fighting it. This guide covers when a voice helps and when it doesn't, how many words fit in 30 seconds, recording on a phone, the rules for AI voices, mixing with music, and captions.

Key takeaways

• A 30-second film holds about 75 spoken words at a calm pace, and closer to 60 if you leave the first and last few seconds to the music; time your script aloud before you record.
• A phone in a small furnished room, held about a hand's width from your mouth, beats a good microphone in a large empty room with hard surfaces.
• ElevenLabs' terms, last updated October 9, 2026, limit free users to non-commercial use, and its policy forbids cloning a living person's voice without their explicit and informed consent.
• When the voice starts, pull the music well down and keep it there until the voice stops; fade it back up for the ending.
• Every voiceover needs captions, both for people watching on mute and for accessibility, and auto-captions need checking for names and materials.

How to add voiceover to an architecture video

To add voiceover to an architecture video, write a script of about 60 to 75 words for 30 seconds, record it on your phone in a quiet, furnished room, and place the recording on its own track under the picture in your editor. Lower the music under the voice, fade it back up at the end, add captions, and export one file. If you use an AI voice instead, use a paid plan that allows commercial use, and only clone a voice whose owner has explicitly agreed.

Before any of that, decide whether the video needs a voice at all.

Does this video need a voice?

A voice helps when the video has to explain something the images can't show: the constraint you solved, the client's brief, why the project matters to a neighborhood. It's useful for a principal introducing a project on LinkedIn, a recruiting video, a short explainer for a planning hearing, or a narrated case study on your website.

No voice is often better when:

• The film is about atmosphere. Light, material and proportion read best with music and silence.
• The audience scrolls on mute. Meta's own Edits tutorial notes that "a large audience watches their content on mute" (Edits by Instagram, as of October 2026). Captions carry the message there, not the voice.
• The script would repeat the caption. If the post text already says "a three-bedroom house on a sloping lot in Asheville," saying it again adds nothing.
• You'd be reading a brochure. "Seamless indoor-outdoor living" spoken aloud sounds worse than it reads.

A middle path: a ten-second spoken introduction, then the project with music only. How long an architecture video should be helps you decide how much room you have.

Write a script that fits the length

Spoken word counts are the easiest way to keep a voiceover from being rushed. At a calm, unhurried pace, about 2.5 words a second is a workable rule of thumb. Read your script aloud with a timer anyway: speaking speed varies a lot from person to person.

A structure that works for a 30-second project film, with placeholders for your own facts:

1. Where and what (one sentence). "[Project name] is a [type] on [site condition] in [city]."
2. The problem (one sentence). "The brief was [what the client needed], on [the constraint]."
3. The move (one or two sentences). "We [the key design decision], so [the result a visitor notices]."
4. The close (one short sentence). "[Completed / in design / under construction], [year]." Or the studio name.

Write for the ear: short sentences, one idea each, no lists of materials, no square footage unless it's the point. Match sentences to scenes: if the third scene shows the stair, the sentence about the stair should land there. For a film whose scene order you don't control, such as an AI-generated one, write the script after you've watched the result, not before.

Recording your own voice on a phone

Your own voice is the most credible narrator for your own work, and a modern phone is good enough if the room is right.

• Choose a small, soft room. A bedroom, a closet full of clothes or a car parked somewhere quiet. Avoid the finished project itself if it's all glass, concrete and hard floors; the echo is obvious.
• Hold the phone about a hand's width away, slightly off to the side of your mouth so your breath doesn't hit the microphone, and speak toward it.
• Silence the phone and turn off notifications, the fridge, the air conditioning.
• Record a few seconds of silence first so you can hear the room before you speak.
• Do three full takes, then pick the best one rather than stitching words together; edits in speech are easy to hear.

Instagram's Edits app helps with two parts of this. Its teleprompter scrolls your script while you record, with adjustable size and speed (Edits by Instagram, as of October 2026), and its enhance tool "cleans and balances your audio by reducing background noise, leveling your volume" (Edits by Instagram). Instagram's Edits app for architects covers the rest of the workflow.

AI voices: commercial use, consent and disclosure

An AI voice can read your script in a polished, consistent voice, or in a cloned version of your own. Two sets of rules matter: the tool's terms and the platform's.

The tool's terms. Using ElevenLabs as an example, as of October 2026:

Other voice tools have their own terms; read them for the same four questions. If you clone a staff member's voice, get their consent in writing and agree what happens to the voice model if they leave the firm. This is a summary, not legal advice.

The platform's rules. YouTube requires disclosure when AI makes "a real person appear to say or do something they didn't do," but lists "cloning one's own voice to create voice overs or dubs," caption creation and "using effects to enhance previously recorded audio" among the uses that don't need it (YouTube Help, as of October 2026). Meta announced in 2024 that it would "require people to use this disclosure and label tool when they post organic content with a photorealistic video or realistic-sounding audio that was digitally created or altered" (Meta). A synthetic narrator presented as a real member of your team crosses that line; a voice you've cloned from yourself, reading your own words, generally doesn't on YouTube. When in doubt, say so in the caption.

Mixing the voice with music

Most narrated project videos fail here: the music is as loud as the voice, and neither wins.

1. Put voice and music on separate tracks. If the music is already mixed into your video file, that file's audio is the music track.
2. Leave the opening to the music. Start the voice two or three seconds in, after the first image has registered.
3. Pull the music down under the voice. Start by lowering it by around 10 to 15 dB while the voice plays and adjust by ear; it should feel too quiet in headphones and right on a phone speaker.
4. Fade, don't cut. Bring the music down over half a second before the first word and back up over a second after the last word.
5. Give the ending back to the music. The final scene and the end card play with music alone.
6. Check on a phone speaker at medium volume, then on earbuds. If you can't understand every word on the speaker, lower the music further.

If your film came from Arch2Video, its soundtrack (Still Air by default, or Quiet Piano or Concrete) is mixed into the MP4. Quiet Piano sits under a voice more easily than a fuller track, and because you can switch the soundtrack and re-export without paying again, you can also choose none and add your own licensed music under the narration. Whatever you use, a studio account posting its own work is commercial use; music for architecture videos covers which tracks are cleared.

Captions for every voiceover

A voiceover without captions is lost on anyone watching muted and inaccessible to anyone who can't hear it. Burned-in (open) captions are the safest choice for Reels, TikTok and Shorts; on YouTube and your website you can also upload a caption file. Edits generates captions from your recording automatically, with an editable transcript (Edits by Instagram); most editors offer something similar.

Check every caption for names, places and materials, keep lines short, and place them a little above center so they don't sit under the platform's buttons. Accessible architecture videos explains caption and transcript requirements for US audiences.

Tools for each step

Feature names are from each tool's own pages as of October 2026, including Meta's Edits tutorials, Blackmagic's product page for Resolve (Blackmagic Design) and Canva's help page on audio tracks (Canva Help Center). For a vertical Resolve project, editing vertical video in DaVinci Resolve covers setup and export.

Frequently asked questions

How many words do I need for a 30-second voiceover?

About 75 words at a calm pace, and about 60 if you leave a few seconds of music at the start and end, which usually sounds better. Read the script aloud with a timer before recording, because speaking speed varies; if you're rushing to fit it in, cut a sentence rather than speaking faster.

Can I use an AI voice for my firm's Instagram videos?

Yes, if the tool's terms allow commercial use on your plan and you have the rights to any voice you clone. With ElevenLabs, for example, free accounts are limited to non-commercial use and paid plans allow commercial use, as of October 2026, and cloning a living person's voice requires their explicit, informed consent. Avoid presenting a synthetic voice as a real member of your team.

Do I have to disclose an AI voiceover?

It depends on the platform and how realistic the result is. YouTube requires disclosure when AI makes a real person appear to say something they didn't, but not for cloning your own voice for voiceovers or for enhancing recorded audio. Meta has said it requires its AI label for organic posts with realistic-sounding audio that was digitally created or altered. If you're unsure, disclose it in the caption.

Should the music stop when I start talking?

Not completely; dropping it to silence makes the cut obvious. Fade it well down just before the first word, keep it low and steady under the voice, and bring it back up after the last word so the ending plays with music alone. Test on a phone speaker: if every word is clear there, the balance is right.