Skip to content
Technology

The best models, our own method.

Video Factory doesn't reinvent AI models: it orchestrates them. Its value lies in the prompt engine, shot detection, URL-based editing and the safeguards that make production reliable.

The pipeline

Five steps, three of them paid.

Only the script, the storyboard and the video call on billed models. Shot detection, editing and every change are computed without generative AI.

BriefAudience, place, message
Script12 scenes + captions
Storyboard12 reference frames
Video15 s, 9:16
EditShots, logo, music
PublishingScheduled, 4 networks

Script and captions

A latest-generation language model receives the brand kit and the brief, and returns a structured object: topic, 12 scenes, image prompt, video prompt, three captions.

Storyboard

A latest-generation image model produces a vertical storyboard of twelve frames. The brand's visual identity is injected as the top priority.

Video

A state-of-the-art video model animates the storyboard into 15 seconds in 9:16. The result comes back through a signed notification, with a fallback check.

Shot detection

A cut is a jump in the image.

Video Factory extracts a thumbnail every quarter second, measures how much it differs from the previous one, then calculates a threshold specific to each video from the median and the median absolute deviation. Peaks above the threshold become cuts, and shots that are too short are merged.

The result: reliable shot splitting across very different videos, with no manual tuning, and a free rerun if needed.

URL-based editing

No render, just an address.

The final edit isn't a file re-rendered on a server: it's a recipe that describes the segments, slow motion, music, fades, logo and end card. Changing a shot changes the address, and the new version is ready in a few seconds.

  • No AI cost when editingEvery change is free of generation costs.
  • Versions savedEvery saved edit keeps its recipe and its length.
shot 1   0.25 → 1.30 s
shot 3   1.35 → 2.75 s
shot 6   4.35 → 5.95 s   slow motion 24%
music    from the brand kit
fade     3 s out
logo     1.5 s fade in
card     1.5 s end card, black background
Architecture

Where things run.

The app and its database are hosted in Paris. Media is stored with a specialised host. Generations go through the APIs of leading AI providers, listed in our privacy policy.

You and your clientsBrowser, approval link
Video Factory appNext.js, hosted in Paris (Vercel, cdg1)
DatabaseSupabase Postgres, EU region, server-only access
Language and image modelsScript, captions, storyboard
Video modelVideo generation
Storage and editingMedia and edit recalculation
Security and responsibility

Safeguards at every level.

Encrypted keys

Your account keys are encrypted with AES-256-GCM before storage and never sent back to the browser: only a masked version is shown.

Isolated data

The database can only be accessed by the server. Every request checks that the user is allowed to see the client in question.

Passwordless sign-in

A sign-in link by email. No password to remember, so no password to steal.

Signed webhooks

Callbacks from the video provider are signed (HMAC): a forged notification is rejected.

Cost safeguards

Monthly cap, credit deducted at launch, automatic pause after two failures in a row.

Responsible AI

No faces, no real people, no minors, no recognisable real brands or places in the prompts.

Independence

Models change, your method stays.

AI models improve every quarter. Video Factory keeps each provider behind a simple interface: when a better model comes out, it replaces the old one without touching your brand kits, your briefs or your habits.

Your first video, on brand, today.

Create your account in one minute: 3 free videos (2 AI videos + 1 motion video), no card required. Rather have us handle it? Request a demo video made by our team.