First short free. Pick a look after you sign up.
In short
An AI video maker turns a prompt into a video by running several models in sequence. One writes the script, another reads it aloud, others find or generate the visuals, and a renderer combines everything into one file. Your prompt only directly controls the first step, so most of a video's quality is decided there. Viibeo publishes this page and sells an AI video maker. We use our own pipeline as the worked example because we can read its code, and the numbers below come from that code. Other tools order these steps differently, and some generate picture and sound in a single model.
- An AI video maker is usually a chain: script, scene cuts, voice, visuals, captions and render.
- The script comes first, so a vague prompt spoils every step after it.
- In Viibeo, a 30-second video becomes 4 scenes of about 17 to 20 spoken words each.
- Scene length follows the voiceover, not the other way round. This is why a finished video can differ from the length you picked.
- Generated images and stock clips are matched to each line one at a time, so visuals are the step most likely to drift from the script.
An AI video maker turns a prompt into a video by running several models in sequence. One writes the script, another reads it aloud, others find or generate the visuals, and a renderer combines everything into one file. Your prompt only directly controls the first step, so most of a video's quality is decided there.
Viibeo publishes this page and sells an AI video maker. We use our own pipeline as the worked example because we can read its code, and the numbers below come from that code. Other tools order these steps differently, and some generate picture and sound in a single model.
What happens after you press generate?
Here is the whole chain for a faceless short, in the order Viibeo runs it:
| Step | Input | Output | What can go wrong |
|---|---|---|---|
| 1. Script | Your prompt, length, look | Title, spoken lines, shot list | A vague idea gives a generic script |
| 2. Scenes | The script | One shot per 8 seconds or so | A long line holds one picture too long |
| 3. Voice | Each spoken line | One audio file per scene | Odd names and numbers sound wrong |
| 4. Visuals | Each shot description | One image or clip per scene | A picture that misses the line |
| 5. Captions | Each audio file | Word-by-word timing | Text and speech out of step |
| 6. Render | All of the above | One vertical video file | Captions covered by the app's buttons |
Picture-first models such as Runway or Veo take a different route: they draw a clip directly from your description. That suits a single striking shot. A narrated short needs a script and a voice as well, which is the job of a chain like this one. If you are unsure which you need, our guide on how to make an AI video starts with that choice.
Step 1: How does the prompt become a script?
A large language model turns your prompt into a title and a list of scenes. Each scene has the line to be spoken, a description of the picture and a short camera direction. The model also labels the role of each scene, such as the hook or the reveal, and picks a mood for the background music.
It writes to a word budget. Narration runs at about 2.5 words per second, so the length you pick sets how much it may write. Before we added per-length budgets, every scene was told to run 7 to 8 seconds, and in our October 2026 audit a 15-second video came out at 31 to 36 seconds.
Compare two prompts:
| Prompt | What the writer has to work with |
|---|---|
| "Make a video about money" | No audience, no claim, so it writes in generalities |
| "3 money habits people in their 20s regret skipping, for beginners, ending on the one that costs the most" | An audience, a structure and an ending |
The model can also be confidently wrong. If your video states facts, check them in the script before the visuals are made, because this is the cheapest place to fix them.
What if you already wrote the script?
Then the model should not rewrite it. In Viibeo, the "From script" option keeps your words and your scene breaks, and the model writes only the visuals. A note in square brackets, such as [close-up of a cat], steers the picture for that scene and is never spoken.
Step 2: How is the script cut into scenes?
Each scene gets one visual, so the cuts set the pace. Viibeo aims for one shot about every 8 seconds and at least 3 scenes:
| Video length | Scenes | Spoken words per scene |
|---|---|---|
| 15 seconds | 3 | 11 to 13 |
| 30 seconds | 4 | 17 to 20 |
| 60 seconds | 8 | 17 to 20 |
If the writer goes over 24 words in a scene, the pipeline splits it near a sentence end and gives the second half a new camera angle. Long stock scenes of 6 seconds or more also change to a second clip halfway through, so the picture does not sit still. For how this affects watch time, see how to write scripts for retention.
On paid plans, you can read the script at this point, after your first finished video. You can rewrite lines, add or remove scenes, or ask for a new take. If you do nothing, the video continues on its own after 15 minutes. When you rewrite a line, its picture description is written again, so the visual follows your new words instead of the old ones.
Step 3: Where does the voice come from?
A text-to-speech model reads each scene's line and returns one audio file per scene. It reads exactly what is written, which makes spelling part of the performance. "$10k" and "ten thousand dollars" can sound different, and an unusual brand name may come out wrong. On most English and Spanish voices, you can add the word and how it sounds in script review or the scene editor, then listen to it before you generate. Captions keep the real spelling.
The measured length of each audio file sets how long its scene lasts. That is why a "30-second" video might come out at 33 seconds. The pipeline does not cut your words to hit a target. For help picking a voice, see how to choose the right AI voice.
Step 4: How are the visuals chosen?
There are three common sources, and they behave differently:
| Source | How it follows the script | Best for | Trade-off |
|---|---|---|---|
| Stock footage | Search terms from each line, then the best unused clip | Real places, everyday actions | Abstract or very specific lines match poorly |
| AI images | The shot description goes to an image model, with your look | Stories, history, anything unfilmable | Costs more; stills need camera motion |
| AI video clips | The shot description or a still goes to a video model | Motion you can't find in stock | Costs the most and takes the longest |
The word "match" hides most of the difficulty here. A stock search for "the economy is slowing" may bring back someone looking at a graph, which is correct but dull. If the visuals feel random, rewrite the line to name something a camera could film: "shop windows with closing-down signs" gives a far better search than "the economy is slowing".
A reference image you attach keeps the same subject and style across scenes, on the AI image and video paths.
Step 5: How do captions stay in sync?
Timing comes from listening. A speech-to-text model transcribes the finished voiceover and returns a start and end time for every word, which is how the highlighted word lands exactly as it is said.
Speech-to-text can mishear, though, especially on names. So Viibeo takes only the timing from the transcript and takes the words from the script. Your spelling appears on screen even when the transcriber heard something else.
Captions are then grouped into short pages that stop at sentence ends, and they are positioned clear of the buttons and text that TikTok and Reels overlay on a video. The opening title already says the first line, so captions join when that title fades, rather than repeating it underneath.
Step 6: What does the render do?
The renderer draws each frame: picture, camera motion, cuts, captions, title card, progress bar, music and voice. It lowers the music under the narration, then normalizes the finished audio to about -14 LUFS, a common target for short-form video, so your video is not noticeably quieter than the ones around it in the feed.
Free exports are 720p at 30 fps with a viibeo.com watermark. The result is a normal video file that you can download, or post automatically with Autopilot on Starter and above.
Why doesn't the video match what I imagined?
Most disappointing results trace back to one step. Work from the top of this list, because fixing an early step also fixes what comes after it:
- The message is wrong or generic: the prompt lacks an audience or a claim. Rewrite the prompt.
- The facts are wrong: the writer guessed. Correct them in the script.
- A word sounds wrong: add how it should sound under Pronunciation, then regenerate that scene's voice.
- A picture misses the line: rewrite the line in concrete terms, or replace the scene's media in the editor.
- The video runs long: there are too many words. Cut words rather than speeding up the voice.
How much does it cost to make one?
Stock footage is the cheapest path, and AI video is the most expensive, because each generated clip has a real cost. On Viibeo, a free account's 10 signup credits cover one stock-footage short of up to 30 seconds, and the second try of that kind is free. AI visuals start at 33 credits for 15 seconds. Lite costs €17.50/mo billed yearly, and full details are on our pricing page.
If you want to compare free options from other tools, we checked 12 of them in What is the best free AI video generator?.
Try the chain with one prompt
Pick one audience and one claim, and keep it to 30 seconds:
- Draft the idea with the free faceless video script generator, or write your own.
- Paste it into the AI video maker and choose stock footage for the first run.
- Watch the result on your phone with the sound on, then again with it off.
- Note the first step where it went wrong, and change only that.
Keep reading
- How to make an AI videoThe practical walkthrough, from prompt to posted short.
- Text to video AIHow the prompt-to-short path works in Viibeo.
- Best faceless video generatorTools for shorts without a camera.
- Free AI video generatorMake a watermarked faceless short.
Frequently Asked Questions
How does an AI video maker work?
Most run a chain of models. One writes a script from your prompt, a text-to-speech model voices it, stock search or image and video models supply the visuals, speech-to-text times the captions, and a renderer combines everything into one file. Some cinematic tools skip the chain and generate a clip directly from your description.
Can an AI video maker use my own script?
Many can. In Viibeo, the "From script" option keeps your words and scene breaks, and the model writes only the visuals. Notes in square brackets steer the picture and are not spoken.
Why is my AI video longer than the length I picked?
Scene length follows the voiceover, and narration runs at about 2.5 words per second. If the script has more words than the length allows, the video runs longer. Trimming words is the cleanest fix.
Why don't the visuals match my script?
Each line is turned into a picture description or a stock search, and abstract lines give those steps little to work with. Name something a camera could film, such as a place, an object or an action, and the match improves.
Do I need to edit an AI video afterward?
Not always. A chain like this produces a finished file, with captions and audio already mixed. Editing is useful when one scene misses, and it is usually quicker to fix that scene than to start again.
Are AI video makers free?
Some offer a limited free allowance, usually with a watermark. Viibeo's free account covers two watermarked stock-footage shorts of up to 30 seconds and does not renew. We did not find an unlimited, watermark-free option among the 12 tools we compared in November 2026.
Written by
Audiencon
Founder of Viibeo
Builds Viibeo and writes this journal from the product: faceless shorts with script, voice, captions, and auto-post.
