Reelkit gives your coding agent ten looks to build a short video in. This guide says when to choose each one and gives you a prompt to paste. By the end you will know which look fits your video, what the agent will ask you, and where it stops for your approval.
What you need
Five things. The install guide walks through each of them.
A coding agent: Claude Code, Codex, Cursor or Gemini CLI.
Node 22 or newer.
ffmpeg, on your path.
The Reelkit CLI: npm install -g reelkit-cli.
A free account: reelkit auth login.
The skill installs itself the first time you log in or start a project.
The agent does the work and stops twice for your yes. Nothing is built before you approve the plan, and nothing is rendered before you approve the preview.
The agent finds out what the video is about, agrees the look with you and writes a plan: the scenes, the narration and the exact words on screen.
Your turnYou approve the plan.
It records the narration and gathers what the scenes need: it searches the shared library first and generates only what is missing. For a scene that needs footage it can also generate a short video clip. It can also remove the background from a clip or from your own footage.
It builds the motion in Remotion, with library components and sounds where they fit.
It renders preview frames of every scene, looks at them and fixes what is wrong.
Your turnYou approve the frames.
It renders the video on your machine and gives you the path to the file.
Have a video you want yours to feel like? Give the agent the link or the file before it plans. It measures the structure, the pacing and the cuts, and builds a new video in that spirit. Nothing from the reference goes into yours. See the recipe
One look for the whole video: a video that mixes looks reads as a template. Pick the look that fits, open its prompt and paste it into your agent.
Restrained
Calm and expensive: one background, one hero colour, one typeface, and one object that changes shape instead of cutting.
Showreel
Made to stop the scroll: hard cuts on the beat between full-colour backgrounds, type that fills the frame, and a camera that never stops.
Either mode
Works both ways. Tell the agent which one you want.
The prompts are written to the agent, so they are in English on every page. Each one tells it what to ask you first, names the look, and holds it to the two approvals.
1
Product walkthrough
Restrained
For an app, a site or a feature, shown through your real screenshots. One device or browser frame stays on screen and the content inside it changes, with one pointer label at a time.
For results, growth or a comparison: anything with figures. Each scene holds one figure or one chart, the number counts up, and the payoff number gets the biggest move of the video.
For an opener, a launch, or a message of a few words that must stop the scroll. One or two words fill the frame, the colour cuts on the beat, and the whole message lands together at the end.
For a mood, a place, or a story told by a voice. Slow moves across pictures under a film grade and grain, a poster-style title to open, and few words on screen.
For technology, AI or nightlife: anything that should feel electric. A near-black background, one glowing colour, a sign-like title and thin lines that draw themselves.
For a personal, warm or teaching tone. A paper background, a handwriting or rounded face, and hand-drawn arrows, circles and underlines that draw themselves onto the point being made.
For a process, a how-to or a recipe of three to five steps. The steps build one at a time, each with a short label beside the action, and the video ends on a short title over the result.
For a testimonial, an introduction, a creator talking, or social proof. The name and role sit in a lower third, the quote gets a card of its own, and a follow line closes. Every name and quote is yours, word for word.
For when you already have your own video. The footage is the picture and the graphics are guests: one overlay at a time, clear of the speaker's face, with captions throughout.
For an opener or an end card made from your own logo file. The motion builds toward the logo from its own shapes and colours, and lands on it exactly as you supplied it.
These are searches of the shared library, not samples of the finished video.
None of these fits?
Describe your own: the message, what happens on screen, the colours, the length and the shape, and send an example if you have one. The agent takes the nearest look as its base, writes a short brief, and shows it to you with the plan.
Mistakes that spoil a Hebrew video
Hebrew on screen breaks in the same few ways. The skill has its own rules for them, and the agent reads those before it plans. This list is for you: it is what to look for in the preview frames.
What goes wrong
The full stop, the comma or the exclamation mark jumps to the wrong end of the line.
What to do
Every Hebrew text container gets a right-to-left direction. The mark then sits at the left end of the last word.
What goes wrong
A word uncovered by a moving edge reads for a moment as a row of dashes, or as a different word.
What to do
Words enter whole: a short rise, or a slight scale down onto their place. No masks and no wipes.
What goes wrong
Two Hebrew words are joined with a hyphen, or a long dash sits in a headline.
What to do
Two words with a space between them. A comma, a full stop or a new line in place of the dash.
What goes wrong
The renderer wraps a Hebrew line wherever it runs out of room.
What to do
Lines are broken by meaning, and each line is an element of its own that never wraps.
What goes wrong
A typeface with Latin letters only is silently swapped for another one, and the weights stop matching.
What to do
A Hebrew face from the kit, such as Heebo or Rubik: one family for the whole video, in two weights.
One thing does not flip: progress bars, sliders and media timelines still fill left to right, as every player does.
Ready to make one?
Install Reelkit, paste a prompt and approve twice. Or look through the shared library first.