The most common problem with narrating short videos isn't the absence of sound—it's sound and picture drifting apart: the narration is still reading the previous line while the footage has already cut to the next shot, or the picture lingers on a product close-up long after the narration has ended. Voice-over works in the reverse order: look at how the video is edited first, then decide what each narration segment says and how long it runs. PixMind's Text to Speech turns each script segment into an MP3, but whether the narration locks into the visuals depends on the segmentation homework you do before writing.
Open Text to Speech, keep your edited clip at hand, generate and download each narration segment separately, then drop them into your editor.
Key takeaways
- Split the narration by visual segment first, then write; each segment carries one information task, and the footage sets its duration.
- Per-generation character limits: 100 characters for signed-out and free-plan users, 10,000 for paid members; the page keeps only one result at a time, so generate and download segment by segment.
- Trust the generated result for timing: the player bar shows each audio segment's total duration—measure one segment first, then schedule the whole video; no words-to-seconds conversion formula needed.
- Merging narration into the video happens in an external editor; subtitles come after the audio track is final, generated from the finished video with a video-to-text tool.
Split the Narration to Match the Visual Rhythm First
The first move in voice-over isn't typing—it's watching. View the edited clip twice, start to finish: the first pass to note the shot order, the second to mouth along silently and feel how many sentences each stretch of footage can hold. Then split the video by shot function.
For a product clip of roughly 15 seconds, split it into opening, demo, and closing:
| Segment | Visuals | Narration task | Rough duration |
|---|---|---|---|
| Opening | The product lands on the table; the camera pushes in | One line saying what it is and who it's for | ~3 seconds |
| Demo | Three action shots: filling, twisting the lid on, starting | One line per action, making the sequence clear | ~8 seconds |
| Closing | Close-up of the finished result; freeze on the logo | One line with a call to action | ~3 seconds |
Each segment's duration comes from the footage; the narration must fit inside that window, not make the picture wait for the words. Write a long essay first and cram it into the timeline, and forced speeding-up is usually the only way out—the listenability collapses instantly. Reading a 300-character product introduction aloud often takes longer than the entire clip.
Segmenting has a practical side too: generation is capped by character count—100 characters per run for signed-out and free-plan users, relaxed to 10,000 for paid members. Narration segments split by storyboard usually fall far below the cap, one segment per file, which suits generating segment by segment.
Rewrite the Script into Spoken, Listenable Language
In a short video, the audience listens to the narration rather than reads it. Two writing habits matter most.
First, delete information the visuals already convey. If the picture is showing the lid being twisted on, the narration doesn't need "next, tighten the lid"; if the camera has already held on the finished result, the narration doesn't need to repeat "as you can see, it came out very smooth." Narration should only say what the picture can't: the name, the use, and the reason behind each action.
Second, one point per sentence, with punctuation controlling the pauses. A comma is a short pause, a period is a breath; short sentences land on shot changes more easily than long ones and give listeners time to digest.
Take the demo segment. First draft: "This product is simple to operate and takes only three steps: first put the fruit into the cup, then tighten the lid, finally press the start button, and after a few dozen seconds you can enjoy fresh juice." Rewritten: "Three steps, done. Fruit goes in. Lid goes on. Press start. Half a minute, fresh juice ready." The rewrite drops the padding, adds pauses, and each sentence lands exactly on one action shot.
While you're at it, write numbers and model codes exactly as you want them read aloud. If "3 steps" should sound like "three steps," spell it out in words; whether the letters and digits in a model number flow naturally when read together should be judged from the generated result.
Audition, Then Choose the Right Voice
There's no single right voice; swap the voice on the same script and the whole video's character shifts. The selection method is straightforward: take one segment—the opening line works well—and generate it with two different voices, then decide after listening.
Compare on two points. One is content style: product demos usually suit a bright, crisp voice, while explainer content leans on a steady, unhurried tone. Two is clarity: focus on the product name, numbers, and places where words run together—whether a voice avoids swallowing or slurring beats whether it sounds nice.
The voice popup filters by engine and language, and hovering over a voice plays its sample right away; samples only hint at a voice's character—the final call rests on results generated from your own script. Don't assume one engine wins every task; try one or two more and you'll settle it in minutes.
Adjust Text and Speaking Rate with Measured Durations
Guessing won't tell you whether the narration length fits; the workable approach is to measure one segment first, then schedule the whole video.
After generating the opening segment, the page's player bar shows that audio's total duration; clicking download saves an MP3 file named with the pixmind-tts- prefix and a timestamp—drop it into a local player or editing track to double-check. Write down all three segments' durations and compare them with the footage durations in your shot list:
- Narration longer than picture: cut copy first. Trimming optional embellishment and repeated information works better than raising the speed—what you remove is baggage, not clarity.
- Still slightly over: nudge the speaking rate. Ranges vary by engine—the Microsoft engine's slider runs -100% to +100% (default 0), Volcengine 0.8x to 2.0x (default 1.0x), MiniMax 0.5x to 2.0x (default 1.0x). Adjust in small steps, regenerating and relistening after every change.
- Narration shorter than picture: you don't have to slow it down to fill. Leaving a second or two of picture without narration often works better than packing words into every moment.
No "X characters equals Y seconds" formula is given here, because such formulas are unreliable: every voice has a different baseline pace, punctuation pauses and number readings shift the real duration, and the same line read by two voices can differ in length by a noticeable stretch. Instead of memorizing a formula, measure the opening segment and use its real duration as the baseline.
Segmented generation needs consistency: use the same voice and the same parameter set across every segment of one video, or the rhythm breaks. Switching engines resets speed and volume to defaults, so write down your tuned values before making changes; the page keeps only one result at a time—generate one segment, download it, then move to the next.
Place the Narration in Your Editor and Add Subtitles
The MP3s only become a finished video inside an editing tool; this step happens outside PixMind, in whatever software you like. The general flow:
- Create a project and import the footage and the three narration MP3s.
- Align each narration segment with its footage per the shot list, leaving a small gap at shot changes so no audio gets clipped by a cut.
- Listen from the top and check that volume and rhythm flow between segments.
- Export the final cut once everything checks out.
Subtitles come after this step, and they must follow the final audio track. Pasting the script text in as subtitles leaves the timeline out of sync with the final sound: after edits and rate changes, every sentence's start and end points have already moved. Instead, hand the finished video with narration to the video-to-SRT tool, generate a timecoded subtitle file from the finished video itself, then bring it back into the editor to check proper nouns and line breaks.
FAQ
When narration runs long, fix the speed or the copy?
Copy first. Trimming embellishment and information the visuals already cover is enough most of the time; touch the speed only when there's nothing left to cut, and overdoing it makes the articulation rushed and blurs the emphasis. Regenerate after editing and confirm the effect on the player bar's duration readout.
Generate in one pass or in segments?
Split by visual segment. A full long script hits the per-generation character cap—100 characters per run for signed-out and free-plan users, with the excess ungeneratable—and the page keeps only one result at a time, so long scripts have to be split and downloaded piece by piece anyway. Splitting by storyboard works with the limits and maps each segment's duration straight onto the footage, so you only re-tune the segment that's off.
At which step should subtitles be made after generating the voice-over?
After the narration is final and the audio track is merged into the video. Subtitle timecodes must follow the final sound: generate them from the assembled video with the video-to-SRT tool, not from the pre-generation script text. Once you have the subtitle file, bring it back into the editor and check proper nouns and line breaks.
Start with One Opening Line of Narration
You don't need the full script first. Open your edited clip and write only the opening segment: one line stating what this is and who it's for. Paste it into Text to Speech, generate, listen, and check the duration. Once the first segment stands, the rest is just repeating the same loop.
