Label
Small white italic — the thing being measured: «rental cost», «floor area».
AI headlines
Subtitles repeat every word. A headline says the one thing the viewer must take away — and holds the screen while it matters.
Open the editor →AI headlines are text overlays: blocks of two to four words over the footage. Captions Space produces these text overlays from speech automatically, with no manual placement.
The keyphrase stage uses word timings and speech prosody. The model produces a short piece of ad copy for each block. The layout engine uses the bounding boxes of detected faces and objects. The renderer produces video with the headlines burned in.
The headline layer enables the point to land without subtitles. The editor includes a storyboard tab with every block. Manual editing enables changes to text, roles and timing. The pause between blocks helps hold the viewer's attention.
Ready-made headline templates from the editor: real fonts, palettes, plates and entry animation. The card shows exactly what the renderer burns into the video.
Oversized centered type
Oversized centered type
Type on plates
Left rail with a bar
Type on plates
Oversized centered type
Oversized centered type
Tilted stickers
Oversized centered type
Oversized centered type
Left rail with a bar
Broadcast band on plates
Oversized centered type
Broadcast band on plates
Type on plates
Oversized centered type
Oversized centered type
Type on plates
Tilted stickers
Oversized centered type
Captions Space includes 20 headline designs. Every one is customizable: role sizes, case, colors, plates, decor and motion.
Word timings come from forced alignment, so every word on screen has its own moment.
Prosody and the scene graph decide where a block starts and what it is built from — deterministic, no network needed.
The model turns the phrase into ad copy: «RENTAL COST», «NEW WALLS». It is a headline, not a quote.
The layout engine computes coordinates in the upper third, clear of faces and objects. The model never returns a pixel.
Small white italic — the thing being measured: «rental cost», «floor area».
Large white caps — the answer the label sets up.
Oversized colour — the number or word the whole block exists for.
Numbers roll to their final value with an ease-out, so a figure lands instead of appearing.
Words fly in one at a time in the order they are spoken, roughly a word per second, and leave the same way. The screen is never asked to hold a paragraph.
Between blocks the frame goes empty for half a second or more. That pause is part of the format — without it the text flickers and the viewer stops reading.
A block is two or three ragged lines with their own sizes and offsets, not one centred string. It behaves like a title card, because that is what it is.
Price, area, wattage, deadline — the label→value pattern is what this mode is built for.
The offer, the benefit and the condition stay on screen while the voice moves on.
Terms and figures get a visual anchor the viewer can hold on to.
The point survives with the sound off, without a wall of caption text.
Subtitles are for comprehension. They repeat everything, so a viewer who lost the audio — or the thread — can catch up on any word. That is why they win on long-form, on interviews and anywhere accuracy matters more than pace.
Headlines are for retention. They show up rarely, hold two to four words, and force a choice about what actually matters in each stretch of speech. On a fifty-second vertical video that choice is the whole edit.
The formats also fail differently. Too many subtitles read as a wall and get ignored; too many headlines read as flashing and get skipped. The density budget — roughly a word per second, with an empty screen between blocks — exists to keep the second failure away.
You do not have to pick once. Run headlines over the hook and the numbers, subtitles over the explanation, and let the layers hand off. Both are timed from the same alignment, so they never fight for the same moment.
Every block lands in a storyboard tab in the editor. Rewrite the line, switch a word between label, value and accent, drop a block, or turn subtitles back on underneath. Nothing is locked to the model's first answer.
An ordinary text overlay is placed by hand: you type the text, pick the spot and pick the moment. Captions Space builds text overlays from speech — it finds the moment from word timings, writes a short block and places it in the upper third, clear of faces. You can still edit all of it by hand.
Captions Space generates AI subtitles, styles them with designer presets and exports the finished video.
Create captions →No credit card required
Runs in the browser — no install, no credit card, and a free plan with 6 minutes of video a month.