Huasheng AI: one script in, one publishable video out in three minutes

Huasheng AI is a video creation tool from Bilibili that takes the text-in, video-out route: you supply a script, or a recording, and it assembles the footage, the voiceover and the subtitles into a finished video. The vendor's claim is three minutes at the fastest. What it actually removes is not editing skill but the three chores that eat the most time: hunting for b-roll, timing every line, and recording a narration without stumbling.

The barrier in video production was never deciding what to say; it is turning that into pictures. A 300-word script takes ten minutes to write, while finding thirty shots that match it, hitting the length of every sentence and recording a narration without a single gasp can take an afternoon. Huasheng AI hands that stretch to a model: the script is the only input, and shot matching, speech synthesis and subtitle timing are done by the tool.

It offers two routes to a finished cut. The first is automatic shot matching, where the AI picks footage according to what the script says, which suits commentary and explainer formats where images illustrate rather than carry the content. The second is template-based production, where you apply a preset layout, which suits serialised output: the same visual system runs episode after episode, viewers recognise it as yours, and you never redesign from scratch.

The capability that gets overlooked is voice cloning. Upload ten seconds of your own recording and it produces a personal voice close to yours; alternatively you can use one of the built-in public voices. For an account publishing spoken commentary daily, the point is that you do not have to open your mouth every time. Revise the script and the video can ship.

What Huasheng AI does

Automatic shot matching

After you enter a script or upload audio, the AI matches footage to each sentence and assembles a complete video; the fastest turnaround the vendor quotes is three minutes. It suits explainers, listicles and opinion pieces where you do not shoot your own material.

Template-based production

A set of preset templates lets you produce a standardised video quickly. The value is highest for serialised channels: the layout stays consistent, the audience recognises it instantly, and production time collapses to filling in the script.

Voice cloning

Upload ten seconds of your own recording and the AI builds a personal voice close to it; built-in public voices are available if you would rather not clone. Useful for accounts that need one consistent presenter voice without recording a human every time.

Multilingual translation

Built on Index, Bilibili's in-house large language model, it supports real-time translation across nearly ten languages, with the vendor claiming accuracy up to 90 percent. That helps when one piece of content has to reach several language markets.

One-click publishing to Bilibili

The finished video can be uploaded straight to Bilibili, so you skip exporting a file and re-entering the metadata by hand.

How to make a video with it

1. Sign in to huasheng.cn with a Bilibili account; there is no separate account to create.
2. Enter a script in the creation screen, or upload an audio file as the basis for the content. The script decides how good the shot matching is: the more concrete it is, the less the footage drifts.
3. Choose the production route: automatic shot matching if you want the AI to find footage, or a template if you need one consistent layout.
4. Pick a voice: a built-in public voice, or clone your own from a ten-second recording.
5. Preview the cut and check section by section whether the images actually match the words; swap individual shots or adjust the audio if not.
6. Publish to Bilibili directly, or export a copy for your archive first.

Who it suits

• Individual creators with no editing background and no time to learn, who have topics in mind and need a script turned into a publishable video quickly.
• Small and mid-sized teams whose output cannot keep up with the publishing schedule, using the tool to cut the person-days behind a single video.
• Short video operators distributing one piece of content across several platforms and accounts, who need volume, consistency and a uniform layout.
• Knowledge channels in history, finance or entertainment, where narration carries the content and images support it, and automation from script to footage pays off most.
• Students and educators producing teaching videos or course material, where the priority is explaining a concept clearly rather than shooting technique.

Limitations to know

• Public material does not state pricing, free allowance or membership benefits, so confirm the billing model on the official site before putting it into production.
• Maximum video length, export resolution and file formats are not described publicly; projects needing long-form or high-resolution delivery should test first.
• Three minutes is the fastest case. Real time depends on script length, how hard the footage is to match, and the queue at that moment.
• The 90 percent translation figure is the vendor's own claim. Technical terms, personal names, place names and puns still need human review; do not ship it as final subtitles unread.
• It is tightly bound to the Bilibili ecosystem: one-click publishing points there, so other platforms still require an export and a manual upload.
• Voice cloning touches personal voice rights. Cloning someone else's voice requires their permission, and AI-generated content should be labelled as the platform requires.
• Whether the automatically matched footage is licensed for commercial use, and whether it can be modified, depends on the licence of the platform's asset library and should be confirmed item by item before commercial use.

Frequently asked questions

Do I need editing skills for Huasheng AI?

No. The flow is write a script, choose a template or let the AI match footage, pick a voice, preview, publish. There is no timeline work. Writing matters more than editing, because how specific the script is sets the ceiling on the visuals.

Can the video go to platforms other than Bilibili?

You can export it and upload elsewhere, but one-click publishing is aimed at Bilibili. When distributing across platforms, note that labelling requirements for AI-generated content differ between them.

How much audio does voice cloning need?

The vendor states ten seconds of recording is enough to build a personal voice. Ten seconds is a very short sample, so stability depends on the recording conditions; record once in a quiet room and avoid background noise.

Is cloning someone else's voice acceptable?

No, not without permission. A voice is a personality right in the same way a likeness is, so cloning another person's voice requires their explicit consent, and AI-generated content generally has to be labelled.