Ziztu AI YouTube Video Maker：如何逐步製作不露面的 YouTube 影片（2026 更新）

I asked our AI YouTube Video Maker for a film about the size of the universe. It drew the universe. It also drew ship hulls, loose rocks and random planets that nobody asked for.

I build Ziztu, so weigh that. That test is why this week's update to the AI YouTube Video Maker is mostly about restraint: one art style for the whole film, no props the script never mentions, a script you read before anything is drawn, and real photos when a film names a real person. Here is how it works now, step by step, with real screens from getziztu.com.

What it is

You paste a YouTube link or type an idea. You get a narrated film of 3 to 60 minutes for a faceless channel, with the script, the pictures, the voice, thumbnails, titles and a description. The price in credits is shown before the film is made.

The start screen. One box takes a link or an idea. Look and length are chosen right there.

What is new this week

• One-click Generate. Link or idea in, a setup sheet, then the whole chain runs to the script. There is one stop, at the finished script, where you read it and then approve the price.
• Openers that hook. The first lines now follow opener rules: say who, when and where inside two sentences, then pull the viewer in.
• A sample for every narrator. 33 voices, each with a short description and a sample you can play. Roger is the default for history films, and a voice you pick always wins.
• Copy a pasted video's drawing style. Paste a link and the film is drawn in that video's style, with our own guide character, not a copy of theirs.
• Real people shown the right way. Real photos with a credit when the look is photographic, drawn likenesses when it is a cartoon, or both.
• One art style, no added props. Every picture follows the same look, and the pictures show what the script says.

Step 1: Paste a link or type an idea

Two ways in:

1. Type an idea: "Why Taiwan makes the chips", "How Roman roads were built", "The size of the universe".
2. Paste a YouTube link to a video that is working. The tool studies how it is built: the title pattern, the hook, the chapters and the pace. A link only works when the video's captions can be read.

Either way, you can also keep a video's structure and change the subject. That is the useful one for a channel: you take a format that works and tell your own story in it.

Step 2: Press Generate and read the script

Generate opens one setup sheet: the look, the length (any whole number of minutes from 3 to 60), the tone, the narrator, and Basic or Pro with the estimate. Then it plans and writes, chapter by chapter, and stops at the script.

That stop is the point. You read the whole script before a single picture is drawn. Change any line. The tool shows an originality score for how far your words are from the source video, and marks passages that sit too close. Facts the writer is not sure of are flagged, and you can ask for a web check.

No picture is bought before you press "Make the film".

Step 3: Pick a look and keep it

There are 18 looks. One look carries every picture of the film, so a 40-minute video does not drift from one style to another halfway through.

The looks, with what each is good for: Map and cast for the big history explainers, Old masters for empires and wars, Engraving for mysteries, Deep Space for planets and galaxies.

Leave it on "Picked for you" and the tool chooses a look that fits your topic. The realistic look needs YouTube's label for altered or synthetic content, and the tool reminds you when it is time to upload.

Step 4: Check that every shot fits the words

Every shot gets a picture drawn for the line the narrator says over it. Here are six pictures from a short film about the size of the universe, each next to its line:

"The Sun is the next stop, about 150 million kilometres away." The picture is the Sun. No ship hulls this time.

A map is drawn only for the part of the story that is about a place. Data cards and diagrams appear when numbers or a process carry the point.

Step 5: Copy a style, with your own guide

This is the new part I am proudest of. When you paste a link, the film can be drawn the way that video is drawn: the palette, the line, the level of detail. It takes two style frames from the video and uses them as the look.

What it does not do is copy the channel's mascot. The film gets its own guide, a thin round-head figure with dot eyes in one or two plain garments, so it fits the style without being their character.

Our output from a pasted video of a Russian history channel, drawn in its style with our own guide. Maps use real relief.

I tested this several times before it was good. The first try reinvented the guide, drew too much detail, and put flat maps where the source used satellite relief. That is why the fix is in the update.

Step 6: Real people and real places

If a film names a real person, you choose how they appear:

• Drawn in the film's look. Made recognisable by public features. The default for stories, comedy and explainers.
• Their real photo. A freely licensed photo from Wikimedia Commons with a small credit line, shown as it is and never redrawn. The default when the look is photographic, so an AI face never stands in for a real person.
• Both. The real photo when the narrator first says their name, then drawn scenes for the rest of the story. The default for history, politics, news and biography in a drawn look.

On the Plan screen you can change the choice for each person. Countries, cities and buildings can appear as real photos with a credit too.

A story about countries can look up who holds the office today, checked on the web, so the film does not name last decade's leader. You can also add a real person yourself, or add up to 3 photos of someone you have the right to use, and they are drawn in your film's look.

Step 7: Basic or Pro

Both tiers use the same writer, the same narrator voices and the same publish pack. The pictures and the motion are what change.

Basic is the lowest price: AI pictures with camera moves, zooms, maps and data cards. Pro adds AI video clips on the key moments.

On both, Claude Sonnet 5.5 writes the script and ElevenLabs reads it. The price in credits is on the button before anything is made.

Step 8: The publish pack

When the film is done you also get what you need to upload it: title options, three thumbnails, a description with chapter timestamps, tags, a pinned comment and an SRT captions file, with one zip of everything. The topic appears on the thumbnail as a short label.

Thumbnails for the universe film: "Earth is a...", "Size of the universe, moon to edge" and "93B light-years".

Download the MP4 and upload it yourself, or send a watch link that anyone with the link can open.

Honest limits

• A pasted link needs readable captions. If the video has none, paste the idea instead.
• Basic has no AI video clips. It is pictures with camera moves. If you want real motion, that is Pro.
• AI writers can be wrong. Flagged facts are a hint, not a guarantee. Read the script and use the web check on anything you will state as a fact.
• Realistic looks need YouTube's altered-content label. The tool reminds you.
• A film is not a channel. YouTube's rules still apply: it calls repetitive, mass-produced videos "inauthentic content". Using AI is allowed. Publishing videos that could be swapped for each other is the risk.

Try it

400 free credits and no card at getziztu.com/ai-youtube-video-maker. Type one idea, read the script it writes, and only then decide to make the film. If you make one, tell me what it was about.