SceinaSceina
Templates

A template is a
spec sheet for the AI

A document that describes exactly the video to produce — format, identity, structure — and how to edit it, step by step. You give your topic, the AI follows the spec sheet to the letter. No improvisation.

Browse templates

Concretely, what is it?

Imagine hiring a video editor and giving them an ultra-precise brief: "vertical format, 45s, voiceover, black & white palette, an intro then N parts then a call-to-action — and here are the exact steps to follow."

That brief is the template. Except here the editor is the AI — and it follows the brief to the letter. You just give the topic; it does the rest while respecting the spec sheet.

Result: videos that are always different, yet always true to the same mold — your brand, reproducible endlessly.

What the spec sheet requires of the AI

A template isn't a suggestion — it's a spec sheet split into sections. Each one sets a requirement the AI must follow (the order and numbering vary from one template to another — it's the content that matters).

Directives

Mandatory

The AI must follow these above all: create the right format, follow the workflow in order, never make creative choices for you, start the voiceover from second 1. Non-negotiable.

Metadata

Mandatory

The AI must produce exactly this output format: the right dimensions, staying within the target duration. No landscape if the template is portrait.

Brand

Mandatory

The AI must use this identity and nothing else: only these colors, these fonts, this voice. No off-brand color or font.

Inputs

Asks you

The AI must ask you for this info before generating — at minimum your topic. Options (number of parts, voice, logo…) have default values if you don't specify.

Components

Fixed

The AI must place these exact blocks (captions, banners, name-tags) at the intended moments — no other components pulled out of thin air.

Blocks

Reuses

The AI must reuse these assemblies as-is whenever a phase calls for them (e.g. "background + verbatim subtitle") — not reinvent them each time.

Layout

Strict order

The AI must create tracks in this exact order (z-order): banner, component, subtitle, background, voice, music. Otherwise layers overlap incorrectly.

Workflow

The core

The AI must follow the phases one by one, in order, without skipping — each phase calls a specific tool, with checkpoints where you must approve before it continues.

Validation

Blocking

The AI must pass ALL fatal rules before rendering (format, verbatim subtitles, duration…). A single failure means no render.

Pick, brief, rerun

Pick a recipe

Grab a template from the catalog matching your format (news, story, listicle, explainer…). It already carries the format, identity, and structure.

Brief the agent

Just give your topic — "the 5 cheapest countries in Europe." The agent runs everything through its toolkit (~90 tools: research, voice, b-roll, transitions, AI image, render) following the recipe.

Rerun endlessly

Every output is unique but stays true to the format. Rerun with a new topic as many times as you want — perfect for a regular series and a solid brand.

Under the hood: how the agent executes the recipe

Each workflow step is tied to a tool. The agent chains them in order — with approval checkpoints.

  1. 1

    It asks for your inputs

    Your topic, then the options (number of parts, voice, logo…). You provide only what's strictly needed — everything else has a default value.

  2. 2

    It writes the script — you approve

    The agent drafts the script following the recipe's structure (intro → parts → outro). Mandatory checkpoint: you approve before it goes further.

  3. 3

    Voice generated, subtitles aligned

    The narration is generated as a single continuous file, then aligned word by word. The voice dictates the real timing — not durations fixed in advance.

  4. 4

    Cutting, b-rolls, tracks

    It cuts the scenes to match the voice, finds fitting b-rolls, and creates tracks in the order defined by the layout.

  5. 5

    Editing part by part

    Intro, then each part, then outro — placing the recipe's reusable blocks at the right moment, with the right cutting rhythm.

  6. 6

    Validation, then render

    A final check verifies the recipe's rules (format, verbatim subtitles, duration within target…). If everything passes, the video is rendered.

The key point: the recipe's durations are estimates. It's the voice that dictates the real timing, and the subtitles align themselves to the exact words.

Create your own template

Define your format once — duration, voice, visual identity, scene types, transitions. The agent then uses it as the framework for all your future edits. The more you produce, the stronger and more recognizable your brand becomes.

Validated and secured

Every imported template passes a two-step check before entering your catalog — nothing malformed or hostile gets through.

Security check

Anti path-traversal, anti zip-bomb, size and file-count caps. A booby-trapped bundle is flatly rejected.

Structure check

The recipe and its components are checked: the YAML parses, required fields are present, blocks are valid.

Executed by the agent

Once validated, the template is driven by the agent via the MCP toolkit — the same workflow, reproducible every time.

Why templates change everything
  • A format defined once, reused endlessly
  • Every video is unique yet stays true to your brand
  • The agent follows the workflow end-to-end (research → render)
  • Create your own templates — your recipe becomes reusable

Launch your next series

Pick a recipe, give your topic, let the agent edit. One video today, a series tomorrow.

Browse templates