Explainer

An explainer is three beats -- WHAT it is, HOW it works, WHY now -- written as timed shots the studio renders one at a time against the person's elements. Words come after the render: on-screen captions live in the studio's editor, not in the prompt, and the voice is recorded by the person. Nothing here generates speech; a render spends credits only after a yes to its quote.

Before writing

1. Get the three beats from the person in their words: what the thing is, how it works (one action you can SEE), and why now (the problem it ends, the moment it fits). Claims about results, numbers, reviews or awards come from the person; never invent one, and keep claims out of the picture anyway -- they belong in the captions and the voice.
2. elements: the scene must be held to the product or the place. A product explainer needs a photo of the exact product (a prop); a service or a venue needs the place. With none, ask for one: an https image address the person gave you goes in with add_element; a file on their phone or computer comes in through upload_link with purpose reference, element_kind and element_name, so it lands on that element.
3. projects / project when it belongs to a project: the brief says what must always show and never show. Settle the total length (4-30 s; 15-20 s suits three beats) and the frame the first platform wants.

The shape

One scene in write-scene's five parts. Each beat is one timed window holding ONE shot (one camera setup, one clear action, 2-10 s); a beat that needs two actions gets two windows. Write what the camera sees, never what a narrator says: no spoken lines, no on-screen text, no logos drawn in -- the product in the photo carries its own.

• (what) the product or place, still and clear, in its real setting
• (how) hands using it, one action from start to finish
• (why now) the moment it matters: the before, the turn, the after

A worked skeleton (replace everything in it):

Ultra-realistic grounded product video in 9:16. The attached photo of the
Kettle is the exact product; the attached photo of the Kitchen is the exact
place. Style: morning window light, clean true colour, a 50mm frame at
counter height, shallow depth on the product, practical light only.
(0-5s) The kettle on the counter, steam just starting; the camera holds.
(5-11s) Hard cut, closer: a hand lifts it and pours into a cup, the stream
steady, steam rising into the light.
(11-16s) Reverse on the cup: two hands wrap round it as the window behind
goes from grey to sun. End abruptly.
No background music. Only diegetic sound: the kettle's tick, the pour, the
cup set down.
Avoid: on-screen text, logos drawn in, plastic AI sheen, CGI-looking skin,
cartoon reactions, dramatic slow motion, named films, brands or actors.

Show it to the person, then save with write_scene: prompt, seconds, refs with the product first, title, project_id. A longer explainer is several scenes, one per beat, each saved on its own and joined later.

The render, then the join

1. quote prices the scene in credits, beside the balance: keyframes (one still per shot, with the credits for each) and clip (one render per timed shot, each with its credits and a token, and the total for all three shots above). Say the quote's credits and WAIT for a yes in chat.
2. On the yes, approve with scene_id, what "clip", the quote's tokens and the renderer choice it priced (provider, model, frame, duration): every shot without a clip yet, at exactly the quoted credits. what "keyframes" draws the stills first, on their own yes, to see the look before the clip. A token works once, for one hour; a stale quote is refused. affordable: false, or insufficient_credits (needs, available), means stop: credits are added in the studio.
3. job with wait until done; scene shows keyframe, clips, rendered.
4. Several scenes become one video with assemble_clips: their gen:<id>s from library in beat order, transition cut or crossfade, and an optional music bed that is an audio asset:<id> uploaded in the studio's editor. Free; it starts a job, so poll job with wait. The result is a new gen:<id> and an editable cut in the editor.
5. The captions (the name, the claims, the offer) and the recorded voice go on in that editor. Say so; never write them into the prompt.
6. One copy per platform with reformat: 9:16 for vertical feeds, 1:1 or 4:5 for square ones, 16:9 for wide. fit pad keeps the whole picture; crop fills the frame. Each copy is a new library item; share_link hands one over as a download.

Do not

• Never write dialogue, a voice-over or on-screen text into the scene.
• Never state a benefit, a number or a review the person did not give you.
• Never pass a URL as a ref; refs are ids exactly as elements lists them.
• Never call approve without the person's yes to that quote in this chat, or for more than it priced; never point them at a page for credits.
• Never say a render or a join has finished: read scene or job.