Add one skill and your AI agent makes a host-led promo video for your website

What you get

Making an explainer video for your product or website is hardest when it comes to getting someone on camera, plus the filming and editing. The ai-host-video Skill on this site hands that work to your AI agent. Install it, give it a website address, and it reads the page, writes the promo copy and script, and generates a host who speaks with lip-synced mouth movements on the video platform jqsp.com. It adds a funny background extra, B-roll, captions and a QR code, and finishes with a second-language voiceover. Four prompts you can copy and paste are included below.

Let's do the math first. A bilingual video of about two minutes costs about 170 credits to generate, which is about $0.42 at jqsp's plan price. Editing and voiceover run on your own computer at no extra charge. The Skill handles the process. You decide who the video is for and what it should say.

Here is the order of things. First, install the Skill in your AI agent and connect it to jqsp. Next, build the host and the background extras, pick a background, and bind a voice to the host. After that, for each video, you only say the video type, who it is for and which website to promote. The rest follows the steps written into the Skill.

The screenshot above comes from a video made with this Skill. The host and both background extras are AI-generated. All the tutorial videos on our platform, including the video for this article, are made this way.

What the Skill handles

A Skill is a set of instructions written for an AI agent. If that is new to you, see the Skill handbook. ai-host-video turns the places where things go wrong or waste money into firm rules, so you don't have to spell out each one:

Quote the price before spending: Every paid step is test-priced first, and the Skill tells you how many credits it will cost. It submits only after you agree.
Keys stay out of the chat: You save your API Key on your own machine, and the Skill reads it only from there.
No made-up numbers: Prices and figures in the script come only from what is written on the website.
Check every frame: If a storyboard draft shows someone with an extra hand, that one frame is redrawn before rendering, so you don't pay to render a bad one.
Build once, reuse: The host, background and voice are created once and reused in every video after that.

Look at the cost first

Here is what one video of about two minutes actually cost us (jqsp pricing, about $0.0025 per credit at the plan price):

StepCreditsApprox. costHost, three shots (15 + 15 + 10 seconds, 480p, lip sync)160$0.40Three storyboard grids3Less than 1 centSplitting dialogue when adding shotsabout 61.5 centsCaptions, B-roll, editing, second-language voiceover (local)00Totalabout 169about $0.42

For comparison, MiniMax's official site lists public prices for the same family of video models at $0.025 to $0.05 per second at 480P. A 40-second host shot would cost $1 to $2 there. On jqsp it is about $0.01 per second, two and a half to five times cheaper. Prices change, so check both sites for current quotes.

Step 1: Install the Skill and connect the video platform

First, sign up at jqsp.com, open Settings → API Keys, create a new key, check the three permissions read, write and charge, and set a daily credit limit so you can't overspend. Then send this to your AI agent:

Install the content at https://cwyz.com/api/skills/ai-host-video/raw as a Skill.
Save it in your own global skills directory. For Claude Code, that is ~/.claude/skills/ai-host-video/SKILL.md. For other agents, put it in the directory where they read Skills.
After it is installed, follow the "Preparation" section of the Skill:
1. Tell me which file on this machine to save the jqsp API Key in, and how to make it readable only by me. I will fill it in myself. Do not ask me to paste it in the chat.
2. Check Node.js, ffmpeg and HyperFrames. If anything is missing, ask me before installing it.
3. Check the jqsp balance and the video model prices once, then tell me in three sentences which rules this Skill makes you follow.

The last sentence checks that your agent really understood the Skill. It should mention "quote the price before spending," "no made-up numbers" and "check every frame for hand errors."

Step 2: Create the host and background, then bind a voice

Do this only once. The host can be based on your own photo, or the platform can generate a look for them:

Following the ai-host-video process, create a video project on jqsp, then:
1. Host: [use my photo / generate a man around 30 with glasses and a gray jacket].
2. Background: [for example, a library with a green desk lamp and tall bookshelves].
3. Background extra: [for example, a man with curly hair and glasses], who only acts and does not speak.
4. Bind a [male voice / a voice I already uploaded on the platform] to the host.
After it is built, show me the images of the three characters. I will use this project for all future videos.

To use your own voice, first upload a voice recording on the jqsp website to create a voice from it. Don't use photos of celebrities or other people as characters.

Step 3: For each video, say only three things

From now on, when you make a video, answer only three questions: what type of video, who it is for, and which website to promote. The Skill reads the site itself, writes the copy and script, and lists the B-roll material:

New accounts get 10 free credits, plus 1 more for each daily check-in. Spend 1 credit to unlock and it stays open for you.

Why it's written this way: the Skill already covers rules like "no made-up numbers" and "quote the price before spending," so here you only add what it doesn't know: who the video is for, what it promotes, and what style the extras should have. The last line makes you review the script before spending credits. Editing lines is free, but changing the video after rendering means paying again.

Step 4: Approve the script and wait for the final cut

If the script looks right, just reply with one line:

The script is fine. Continue with the Skill's process: check each storyboard frame, report the total price before rendering, organize the B-roll while it renders, and finally give me the final video in both languages, a preview version, the upload title, description and tags, and how many credits each step actually used.

Rendering takes 10 to 20 minutes, and the Skill has your AI agent organize the B-roll and post-production in the meantime. In the English version, the lip sync matches. The Chinese version uses voiceover. If you want lip sync in Chinese, change the language to "Chinese first" in Step 3.

Notes

Label AI-generated content when you upload: The host, the extras and the second-language voice are all AI-generated, and YouTube and similar platforms ask you to tick "AI-generated content."
Only use faces and voices you have the right to use: Don't use photos of celebrities, famous people or other people as characters.
Set a daily credit limit: The Skill quotes the price first, and a limit on your key adds another safety net.
Lip sync works only for the main language: The second language is a voiceover, which suits explainer videos but not close-ups that need matching mouth movements.