Video with AILesson 55 of 66

Contents

Video with AI

Lesson 55 of 66. Sign in to keep your place.

Module 6 · From generations to a finished piece

Real type, added afterwards

8 min read

The rule and its reason

Any text that has to be readable is added in post. Not because models are bad at text in some vague way, but because a letterform is a symbolic shape below the encoder’s preserved resolution, reinvented each frame, and a letter that shifts by two pixels per frame reads as melting even when each individual frame is acceptable.

A title in your editor is rendered from a font. It is identical in every frame. There is no mechanism by which it can drift.

So: shop signs, packaging, screen content, subtitles, lower thirds, end cards. All of it goes on afterwards.

Titles on a generated background

Three practical points that make the difference between a title that sits in the shot and one that sits on top of it:

Contrast. Text must be readable against the busiest frame it covers, not against the frame you happened to pause on. Check the whole shot. If contrast fails anywhere, add a subtle shape behind the text, or a soft gradient, or darken that region of the picture slightly.

Motion. A locked-off title over a moving background reads as an overlay. Matching the title’s movement to the shot’s (a slight drift in the same direction, a scale that follows a push-in) integrates it. This is one line of keyframes and it is what separates broadcast titling from a caption.

Safe margins. Keep text within roughly ninety per cent of the frame, and further in on vertical video where interface elements cover the lower third. Check the actual platform.

Replacing text inside a shot

When the generated sign has to say something specific:

  1. Generate the shot with the sign blank, or with any text, and note that you will replace it.
  2. Track the sign’s surface in the editor. Resolve’s Fusion page does planar tracking in the free tier; Blender’s tracker is free and capable.
  3. Attach your real text to the track, corner-pinned to the surface.
  4. Match the grade, then add grain to the text so it shares the surface texture of the rest of the frame.

Step four is the one that makes it work. Clean vector type over grained footage announces itself immediately.

Generate the plate with a trackable feature: an edge, a corner, a mark. A featureless surface cannot be tracked and you will discover this after paying for the clip.

Fonts, and the licence you did not read

Font licences are real and they catch freelancers regularly. A font free for personal use is not free for a client’s advertisement.

Safe free sources:

  • Google Fonts. Open licence, commercial use permitted, enormous range, and strong coverage of Indic scripts: Noto covers most writing systems, which matters if you are working in Hindi, Bengali, Tamil or Telugu.
  • The Open Font Library and Fontshare for more distinctive faces.

Check the licence file that ships with any font you download. Keep a copy in the project folder, for the same reason you keep the music licence.

Subtitles, which are not optional any more

Most social video is watched without sound. Subtitles are not an accessibility afterthought; they are how the majority of your audience receives the words.

The free route: generate a transcript with Whisper, which is open, free, and runs locally or as a Hugging Face Space, then import the resulting SRT into your editor and style it. Resolve, Kdenlive and CapCut all import SRT.

Always read the transcript before you use it. Automatic transcription gets names, technical terms and accented speech wrong, and a confidently wrong subtitle is worse than none.

The label card

The piece of typography this course keeps returning to.

Several countries now require synthetic media to carry a label a viewer can actually see. Metadata labels and invisible watermarks are real and worth keeping in the file, and both are fragile: a screenshot destroys them and many platform re-encodes strip them. The label that survives contact with the internet is the one burned into the frame.

So put it in the video: a title card at the start, in frame, before a viewer has had time to believe anything. Legible, on screen long enough to read, not hidden in the corner at four pixels tall.

It costs you two seconds and a text layer. The module on law and disclosure deals with what the rules currently ask for and how much they differ by country; the craft point here is that it belongs in the titling pass, with everything else you want a viewer to read.


Today: build your label card as a reusable title in your editor and save it as a template. You will use it on everything.

Every lesson here is open. Your progress is recorded once you pass the Editing Video exam.

Check your understanding

Why is a title added in the editor immune to the melting that affects generated on-screen text?

Pick the one you would defend. Your answer stays private.

Next lesson · 8 minSlow motion is where fake physics goes to be caught

Stuck on something?Ask Chiamaka, your AI tutor

How to cite: Addaly (2026) ‘Real type, added afterwards’, in Video with AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-video/titles-text-and-the-label-card (Accessed: 13 October 2026).