The step editor

Retype a beat’s narration, cut a beat out of the video, move the camera, mask a region, lay audio under a beat — then render a new version.

Open a finished demo and press Edit. The video sits on a stage with its beats laid out on a timeline underneath; you change what you want and press Render, which makes a new video beside the old one.

Editing is free, on every plan, forever.

The five things you can change

Axis	What it moves	Where
A beat	Narration, trim, camera, masks, sound	The timeline and the inspector
The look	Backdrop, framing, cursor, captions	The bar's Look panel
Finishing	Intro, outro, replacement soundtrack	The bar's Finishing panel
The voiceover	Whether the demo speaks at all	The bar
The seams	What happens where a stinger meets the demo	A stinger block

Each is independent, and any one alone is a different video worth rendering. Changing only the soundtrack moves no beat and changes no pixel of the plan, and it still ships.

Narration

Click a beat, retype its line. The field shows the word budget for that beat and what exceeding it costs — the beat stretches to fit your line, up to a point, and past that point the audio starts to drift into the next beat. The hard stop is 140 characters, which is exactly where the beat stops stretching.

Nothing you type is trimmed. The budget is a guard on a generator, because a model cannot be told "no"; a person can be shown the cost and trusted with it.

Trimming a beat out

Mark a beat hidden and it leaves the video. It does not leave the plan: the browser still performs it during capture, and only the picture is cut.

That is not a technicality. "Click inside the modal" only resolves because "open the modal" ran first, so removing a step does not shorten a video — it breaks a replay ten minutes later inside a container. For the same reason there is no reorder and no delete anywhere in this editor. Hiding is the whole cutting primitive.

The trim is a re-timing, not a mask: the surviving beats slide earlier by exactly what was removed, and every narration line, caption cue and seek position moves with them. An edit that hides every beat is refused while you are there to be told.

The camera

Three choices — rest, focus and close — and the payload carries the NAME. The scale each one means is looked up on the server from the design tokens, so there is nowhere to write "4.2×" and no way for one demo to zoom differently from another.

Two behaviours worth knowing:

Not choosing leaves the stored zoom alone. Discovery writes scales that no choice names, and a payload that always stated one would round every such beat to the nearest name on any edit — the editor quietly rewriting a plan nobody looked at.
It is only offered where it can take effect. A beat that clicked nothing has nothing to aim at, so the control is hidden, and a pick that arrives anyway is refused rather than silently ignored.

The camera is not previewed. The zoom is baked at the scale it was captured at, and cropping the stage would invent pixels; the pick gets a chip naming the real scale instead.

Masks

Up to 4 rectangles per beat, each with a treatment: light, heavy and solid. They are drawn on a schematic pad in the inspector rather than over the stage, because the stage's zoom is already baked into the frames and an honest overlay is therefore impossible there.

A mask is drawn INSIDE the camera transform, so it zooms and pans with the video and stays over the thing it is hiding.

Per-beat sound

A beat may layer up to 4 audio clips out of your library, each with its own window into the source file and its own level: quiet, standard and loud. They start where their beat starts, mix UNDER the narration, and fade in and out.

Dragging an edge trims the clip; nothing moves it. A clip's position is its beat's position.
It never lengthens the video. Every bed is cut off at the end of the picture, so a five-minute track cannot buy four minutes of frozen frame.
A hidden beat's sounds never play, with no special case anywhere — the cues are joined to the beats that survived the cut.
The replacement soundtrack in Finishing still replaces all of this. It runs after everything else, by design.

The voiceover switch

Off drops the generated narration entirely and keeps the words: the lines stay in the plan, the field stays editable, and switching it back on brings the same demo back speaking. Nothing is synthesised, so no transcript and no caption file are written.

The beat sounds still mix, which is the point. Muting the voice is how you get an edit whose soundtrack is yours.

What the preview will and will not do

The one law of this stage is that nothing in it may lie. Concretely:

Typing draws the caption plate, exactly — and only where your line differs from the one already burned into the video, or you would be reading it twice.
A trimmed beat is skipped in playback, exactly.
The look is approximate and says so on screen. The backdrop, padding, radius and shadow are redrawn in CSS around the recorded card; the outro card and the baked cursor cannot be shown at all, so the panel says they appear on the next render.
Finishing plays the real files, with the audio sync stated as approximate.
The camera and the seam transitions are named, not performed.

Rendering

Press Render and an edit job is created — a new row, a new video, a new share link if you want one. The source is never touched: the original stays in your library, stays shareable at the link you already gave out, and "undo" is opening it and editing again.

An edit is itself editable, re-renderable and refreshable, with no special case.