The bottom fifth of your video belongs to the app

Why the text is a separate pass at all

It is tempting to ask the model for a film with your price already in it, and it is the one thing worth never doing. Generative video models render letterforms as texture rather than as language: they produce something that looks like type at a glance and falls apart when read, with the second half of a word melting into a shape that is almost a letter. So the instruction goes the other way. The prompt that produces a Nitaris film explicitly forbids on-screen text, captions and subtitles, and every word you want a viewer to read is drawn on afterwards by a normal typesetter over the finished picture.

That is a constraint worth understanding rather than working around, because it is also the thing that gives you control. Type that was generated is baked into the picture at the moment it is made and cannot be moved, corrected or removed. Type that is a separate pass can be placed, and placed differently for each shape you deliver.

The part of the frame you do not own

A vertical video does not reach the viewer as a rectangle. It reaches them inside an app, and the app spends the edges of that rectangle on itself: a caption, an account name, a row of buttons up the right-hand side, a progress bar. None of that exists in the file you uploaded. All of it exists on top of the file when someone actually watches it.

This is the single most common way burned-in text is wasted. A headline placed neatly at the bottom of the frame in an editor — where it looks balanced, because an editor shows you the whole rectangle — is the headline that sits underneath the caption and the share button on a phone. It was never seen. Nothing warns you, because on your screen it is fine.

Nitaris reserves those bands rather than trusting you to remember them. On a 9:16 cut it keeps the bottom 22% of the frame clear, on a 4:5 cut 12%, and on a 16:9 cut 10%; the sides keep 7% of the short edge, and the top 12% on a vertical cut. Text is placed inside what survives, and the logo hugs a surviving corner. Those are the numbers the layout uses, and the reason they differ by shape is that the chrome does.

What each app covers today, though, is that app's decision and it changes without telling anyone. Treat the reserved band as a sensible floor, not as a specification, and do the only test that is ever current: post one, screenshot it on a phone from the feed, and look at what the interface is sitting on.

One film, several shapes, several safe areas

A film is delivered in up to four shapes — 9:16, 16:9, 1:1 and 4:5 — and their safe areas are not the same safe area. This is why the overlay is drawn per delivered crop rather than once onto a master: the film underneath stays clean, and each cut gets its words positioned for its own frame.

The practical consequence is for anyone doing this by hand in any editor. If you burn text onto a vertical master and then crop that master to a square, you have moved the text as well — usually into the middle of the product, or off the edge. Burn last, once per shape, or keep a clean master you can re-cut.

What to actually write

The fields are deliberately short: a headline of up to 60 characters, a price of up to 16, a call to action of up to 24. Those limits are not storage; they are what survives being read at arm's length, at a phone's size, in about a second and a half, usually with the sound off. A long headline shrinks to fit and then stops shrinking, at a floor of 55% of its intended size, because past that it has stopped being a headline and become a caption nobody reads.

A useful test before you type anything: name the one thing a stranger must know that the picture cannot tell them. The picture already shows what the product looks like, how it moves and roughly what it is. It cannot say what it costs, what it is called, or what to do next. Write that, and nothing else.

• A headline earns its place when the product is unfamiliar or its benefit is invisible — what it is, or the one claim a viewer cannot see.
• A price earns its place when the price is the argument: notably low, or notably high in a way that signals what the thing is.
• A call to action earns its place when the film will be seen somewhere the next step is not obvious. On your own product page, there is already a buy button six inches below.
• A logo earns its place almost always, and costs almost nothing — it is the only element that still works when the video is reposted without you.
• All four at once earn nothing. They crowd each other, shrink to fit, and the viewer reads none of them.

A printed price is a promise with a shelf life

The field worth the most thought is the price, because it is the only one that can become false while the file stays the same. A price burned into a video is frozen into every copy of that video — on your product page, in an ad account, in whatever a customer downloaded or someone else reposted. The day you run a sale, the film is wrong. The day you raise the price, the film is wrong in the more expensive direction, and somebody will screenshot it.

This is not an argument against printing prices. It is an argument for printing them deliberately: on films made for a dated moment — a launch, a seasonal drop, a sale that ends — where the film is meant to expire anyway. For an evergreen product film that will sit on a listing for a year, leave the price to the page, which can be edited in a second and costs nothing to change.

The same logic applies, more gently, to anything else time-bound. "New" stops being true. A discount percentage stops being true. A film without those words stays usable for as long as the product looks like that.

The clean master is the asset

The thread running through all of this is that the picture and the words have different lifespans. The picture of your product is good until the product changes. The words are good until your price, your campaign or your platform changes, which is sooner.

So keep them separable. In Nitaris that is the default — the overlay is a pass over a clean delivered film, and what it draws is frozen onto the job when you queue it, so a brand kit edited while the film renders cannot quietly change what was printed. If you are working in another tool, make the same arrangement by hand: export the clean film once, and treat every version with words on it as disposable. You will re-make the words several times. You should only have to make the film once.