- Home
- Skills
- Image & Video
- AI Video Studio: Topic to Finished MP4
More screenshots
Works with the AI Agents you already use
AI Video Studio: Topic to Finished MP4
Most AI agents can write about a video, but they can't hand you a finished MP4.
$9.99
AI Video Studio: Topic to Finished MP4
Example session with this skill installed
/make-video Make a 45-second vertical video about how I make videos with a coding agent.
- Read your context and instructions
- Compiled the ai video studio
Result: a 46.1-second vertical video, finished in 23 minutes including approvals. Render time: 1 min 46 s on an 8 GB laptop.
| Files | edit/final_shorts.mp4 (the same settings suit Shorts, TikTok and Reels) |
| Picture | 1080×1920, 30 fps, H.264, 4.6 MB |
| Loudness | −14.2 LUFS, true peak −1.5 dBTP |
| Style | "Source view", a new style it designed: each beat types in as code, then springs out as big type |
| Voice | Local Kokoro voice, chosen by ear from three auditions |
| Music | Composed for this video, in a minor key at 102 bpm |
The steps it stopped for:
-
Permission to install tools, with the size stated first (about 720 MB, plus 480 MB of models on first use).
-
Platform, audience and brand.
-
The script, 10 lines and 121 words:
This video started as a web page.
Nobody opened a video editor. I typed one sentence to a coding agent.
The agent has eleven skills, plain instruction files that teach it each step.
It writes the script, then stops and waits for my approval.
It generates this voice on my own computer and measures when every word lands.
Those timings drive every animation, and even the music it composes.
Then it renders the video and checks its own frames for mistakes.
The one thing it can't do is hear, so the final check is still my ears.
Want a line changed? It redoes only what that line touches.
What would you make if your next video started as a web page? -
The style, the voice, and a 19-beat storyboard.
-
The go-ahead for the full render, with a time estimate.
What it caught by itself: the first build showed every scene at once. It read the error, fixed the cause, and checked the frames again. It also tightened the opening frame and closed short gaps at nine cuts before rendering.
What it told me at handover:
- The frame check found 0 blank frames in the whole video.
- It can't hear, so it asked me to listen, and flagged one line where speech recognition heard "at" instead of "it".
- Which parts on screen are real project data (the prompt, skill names, waveform, word times) and which are illustrations.
- The choices it made for me, and how long each kind of change would take.
Next steps it offered: re-voice a line (under 2 minutes), swap the music (about a minute, no re-render), or change on-screen text and colours.
Connects securely to your tools. The creator never sees your data.
About this skill
Most AI agents can write about a video, but they can't hand you a finished MP4. AI Video Studio gives your coding agent a complete production pipeline, 11 skills it follows step by step, to turn a topic, script or article into an upload-ready video. Every visual is written as code, the music is composed on your machine, and the free voice runs locally, so there are no per-video fees, no stock music licences and no image-model costs.
In our test run, a fresh session made a 46-second vertical video in 23 minutes, including the time to approve each step. That run had the voice and speech models already downloaded.
How it works
- Checks your machine and asks before installing anything.
- Asks for the brief and your brand, and sets the frame, length and hook for your platform.
- Writes the script and stops until you approve it. It's instructed never to invent facts and to flag anything unverified, but check claims yourself before you publish.
- Picks a style: one of six presets, or a new one designed for your video.
- Makes the voice: a free local voice, your own recordings, or narration from any service.
- Plans a storyboard, checked against variety rules so it never becomes "same card, same fade", and stops for approval.
- Builds the motion graphics, with every animation cued to the spoken words.
- Composes original music shaped to your video's structure.
- Renders and checks its own work: it scans for blank frames, looks at frames from the whole video, and normalises loudness.
- Hands over honestly: what it measured, what it can't verify (it can't hear), and which choices it made for you.
Why this beats prompting it yourself
A video only works when every visual lands on the right word. Prompting by hand can't keep track of that. This pack keeps one timeline.json, built from the real length of each spoken line, and every animation is cued from it. Change a line and the agent regenerates that line's audio, and the visuals follow without hand-editing a single timestamp. The same file shapes the music.
What you get
- Six ready-made styles: Kinetic typography, Editorial collage, Newspaper, Keynote dark, Retro desktop, Manga panels. More come in free updates.
- A style creator that designs a new look for your brand or idea, and saves it for later videos.
- Exports for YouTube, Shorts, TikTok, Reels, LinkedIn and X.
- Word-timed captions built into the design.
- Carry-over: your first video takes the longest. After that, your tools, brand, styles and preferences carry over from
~/.video-studio/, plain files you can read and edit.
Requirements
- A coding agent that can run terminal commands and write files. Built and tested with Claude Code. Chat-only assistants can plan and write scripts but can't render.
- ffmpeg and Node.js 22+. Python 3.10+ for the free voice, music and precise word timings. The
video-doctorskill checks these and gives the install command for macOS, Linux or Windows. - About 1.6 GB of disk on your home drive, once, shared by all your projects (820 MB without voice and music).
What it doesn't do
It doesn't generate photorealistic images or video: visuals are made in code. It doesn't edit camera footage. It's designed for videos up to about 3 minutes. Without ffmpeg or Node 22+ it can only make an HTML preview, not an MP4.
- Visuals are made in code: no photorealistic images or video.
- It doesn't edit camera footage.
- Designed for videos up to about 3 minutes.
- Full setup downloads about 1.6 GB (820 MB without voice and music).
- It's a long agent session, not a quick prompt: a 45-second video took 23 minutes in our test and uses a meaningful share of an agent plan's usage. The test used a top-tier model at high effort; a cheaper model may give different results.
- Renders are CPU-heavy. On 8 GB of RAM, use 2 workers and close other apps.
- The agent can't hear audio. It measures voice and music, but the final listen is yours.
- Built and tested end to end with Claude Code. Other coding agents that read SKILL.md should work, but haven't been tested end to end.
- The free local voice is good but recognisably AI. Check each platform's rules on labelling AI audio.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 9 days ago
Requires
Permissions
File Scopes
Nothing is installed without asking. Network access is used at setup to download the renderer and models from npm, GitHub and Hugging Face, and fonts from Google Fonts. After that it runs offline, apart from new fonts and any web research. API keys are optional and are never printed.