AI video generator

The model writes, the engine builds

This is the split worth understanding before you judge the output. The model chooses a template, orders the scene roles, writes every line of on-screen copy and the narration, and sets how long each scene holds. It never draws anything. Layout, camera moves, entrance timing, colour and the render itself are ordinary deterministic code, which is why two runs of the same brief look like the same product rather than two different apps.

Three types, three different jobs

Promotional writes an ad: a hook, the value, some proof, and a closing ask. Tutorial teaches instead of sells, one scene per step, and refuses a sales ending. UGC writes a creator talking to their phone camera, cut with quick proof beats. The type is not cosmetic. It changes which of the 50 scene roles are even offered, how long a scene may run, and which narrator voice you are allowed to get.

What you feed it besides words

Up to eight images and five documents ride along with the brief. Images are inspected before anything is designed: a vision pass reads what each one shows, notes whether pictures belong together, and reads the real pixel dimensions out of the file header so a tall phone screenshot lands in a tall slot rather than being cropped square. Documents are summarised for context, not transcribed, so a PDF informs the script without becoming the script.

Everything is still editable afterwards

The result opens in a builder with a scene rail, a timeline and an inspector. Rewrite any line, swap a picture, add a scene from the library, change the palette, retime a beat by dragging its handle, rerecord narration in a different voice. Nothing is baked until you export, and the export reuses an existing render when the content has not changed.

How it works, in three steps

1. Step 1

   Pick the type first

   Promotional, Tutorial or UGC. It gates the scene roles, the pacing rules and the voice, so switching later means rereading the whole script.

2. Step 2

   Write the brief like a creative brief

   Product, audience, the two or three points that matter, and the exact closing line you want spoken. Vague briefs produce vague videos.

3. Step 3

   Watch it build, then edit

   Generation runs as a background job and is polled while you wait. When the plan lands, the builder fills with scenes you can change.

The full walkthrough with screenshots is in the guide Make a promo video with AI voiceover.

Limits worth knowing

• The brief box takes 8,000 characters. Uploads are capped at 8 images and 5 documents.
• Tutorial videos need the Starter plan. UGC videos need Pro or higher. A free account can still make promotional videos.
• Generation is a background job polled for up to 16 minutes, and it spends one credit.
• Only landscape and vertical are offered. The 1:1 square canvas exists in the engine but is hidden from both pickers.

Questions people ask

Related pages