I Built an AI Skill That Turns a URL or Raw Footage Into an Editable Video Project

AI video tools are getting better at producing output, but many of them still have the same limitation: they give you a finished MP4 and very little control over what happens next.

If one subtitle is too fast, one voiceover sentence sounds wrong, or one product shot needs a longer pause, the easiest option is often to regenerate the whole video.

I wanted a different workflow.

I recently shipped a major upgrade to Timeline Studio, an open-source, browser-based AI video editor. The new workflow can analyze a website, images, or raw footage, create a first cut, and export both:

• a finished video ready to publish
• an editable .timeline project for future revisions

The idea is simple: let AI handle the time-consuming first draft, while the creator keeps control of the final edit.

What the Skill can do

After installing the Timeline Studio Skill, you can describe the result you want in natural language.

For example:

Analyze this website and create a product introduction video.
Use a professional and energetic English female voice.
Target roughly 60 seconds in 16:9.
If login is required, ask me before continuing.
Export the final video and an editable .timeline project.

The Skill can then plan the content, create a narration, generate synchronized subtitles, select scenes, add background music, and apply visual emphasis.

It supports several editing scenarios:

• talking-head and voice-driven videos
• website and SaaS walkthroughs
• product or brand promotion
• image-based videos
• long-form content condensation
• tutorials
• highlights and short-form clips

The editing strategy changes with the content instead of applying the same template to every project.

Website promotion is more than screen recording

For a website introduction, recording a browser session is not enough.

The system first explores the publicly accessible pages and tries to understand the product, information structure, and most important features. If the key experience requires an account, it asks the user for access instead of inventing conclusions from the landing page.

The resulting video can combine:

• stable page recordings
• smooth keyframed zooms
• lines that guide the viewer's attention
• a small number of accurately positioned highlights
• subtitles and narration aligned with the visible action

We deliberately avoid drawing boxes around everything. Too many highlights make the video feel noisy and reduce their value.

Instead, the Skill chooses between three visual treatments:

• Zoom for a core feature, result, or chart
• Guide line for buttons, text, and reading direction
• Frame for a clearly bounded area that genuinely needs emphasis

A typical move is a smooth zoom in, a short hold, and a controlled zoom out. The camera should feel intentional and stable rather than shaky.

Voiceover becomes the timing source

One of the biggest lessons from earlier versions was that forcing narration into an exact duration can make it sound rushed and unnatural.

The upgraded workflow generates the voiceover at a natural pace first. It then uses the real speech timing to determine:

• caption boundaries
• scene changes
• zoom and highlight timing
• music volume automation
• ending space

A requested duration such as 60 seconds is treated as a target, not a reason to damage the delivery. A clear 54-second or 66-second video is often better than an unnaturally compressed 60-second one.

An editable .timeline v3 project

The most important technical change is the project format.

In .timeline v3, voiceover is stored as separate sentence-level audio clips rather than one long, difficult-to-edit file. Each caption can reference its corresponding voice clip.

A simplified representation looks like this:

{
  "caption": {
    "text": "Career planning should not depend on headlines alone.",
    "audioClipId": "voice-001"
  },
  "audioClip": {
    "id": "voice-001",
    "start": 0.8,
    "duration": 4.2
  }
}

This means you can replace one sentence, regenerate one voice clip, or adjust one subtitle without rebuilding the entire project.

The project also preserves scenes, captions, voice clips, music, overlays, effects, and animation settings. The MP4 solves today's publishing need; the project file solves tomorrow's revision.

Fixing playback loops at clip boundaries

The upgrade also addresses an editor playback issue where a segment could repeat near a clip boundary.

Individual media elements are not always reliable as the master timeline clock. Decode delays and small rounding differences can cause time to move backward or make two neighboring clips appear active at the same boundary.

Timeline Studio now uses a monotonic timeline wall clock as the master and synchronizes media against it. Clip activity is evaluated using half-open intervals:

[start, end)

That prevents the previous clip and next clip from both being selected at the exact same timestamp.

Install the Skill

You can install the current release with:

npx skills add MartinDelophy/ai-video-editor@v0.9.1

GitHub:

https://github.com/MartinDelophy/ai-video-editor

Live demo:

https://video-editor.ai-creator.top

The goal is to produce a strong first cut in the time it takes to finish a coffee, without turning the result into a black box.

AI performs the repetitive first-pass work. The creator still owns the timeline, the details, and the final decision.

Feedback and contributions are welcome.