Vizard Agent

How to put your talking head in a round chat bubble

Last updated 2026-09-15 · 8 min read

Ask Vizard Agent to build the chat look from the clean recording rather than over a version that already has titles on it, and to leave real margin around your head inside the circle rather than merely fitting it. It then times the message bubbles to specific spoken lines and checks the chin clearance at full size before delivery.

What is the short version?

A circle is a smaller shape than the rectangle it is cut from, and it takes the corners first. A head framed comfortably in a vertical video will have its chin clipped the moment it is put in a circular window.

  1. Give Vizard Agent the clean recording, not a finished edit.
  2. Ask for margin around your head, not a fit.
  3. Ask for the bubbles timed to specific lines.

What do you need before you start?

The original recording and a sense of what the messages should say. Vizard Agent needs clean source because a circle placed over an edit that already has captions and lower thirds will crop those into fragments around the rim.

What do you type into Vizard Agent?

Say margin, not fit. Asking Vizard Agent to make your face fit inside the circle produces a face that fits it exactly, which means a head touching the rim on three sides — and that, rather than the crop itself, is the thing that reads as wrong when you watch it back on a phone.

Prompt

Put my video inside a clean circular profile frame like a chat video message, built from the original recording rather than the edited version with titles. Scale the picture down inside the circle so my whole head sits well inside it with clear space under the chin. Add small chat message bubbles timed to the lines I say, and some quiet music under the voice.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the sequence on a real talking-head clip rebuilt as a chat video message. Note how many passes go into the circle alone — it is scaled down three separate times, each after looking at the actual frame.

  1. Rebuilds the look on clean source, with no previous lines or titles.
  2. Checks the round frame and the message animation.
  3. Checks the speech is continuous and at a steady level.
  4. Times the chat bubbles to specific spoken lines.
  5. Looks at the frame to measure the margin around the face.
  6. Scales the picture down inside the circle to create that margin.
  7. Confirms the whole face fits after the reduction.
  8. Adds minimal chat icons and checks they fit too.
  9. Widens the circle and mixes the music under the voice.
  10. Reduces the inner video once more so the chin has guaranteed clearance.

Step five is why this takes passes rather than a calculation. How much margin looks right depends on where you sit in the original frame, so Vizard Agent looks at the actual composed frame rather than applying a percentage.

Step ten is the one that finishes the job. A head that fits looks cramped; a head with obvious space beneath the chin looks like a chat message, and the difference is a few per cent of scale.

Step four is what stops the bubbles feeling decorative. Messages that appear on a rhythm read as animation; messages that appear on the line that mentions them read as a conversation.

What does the result look like?

You, in a circle, with room to breathe around your head, and messages arriving on screen as you say the things they refer to. Vizard Agent checks the final clearance at full size rather than at working size, and confirms the speech runs continuously underneath the music without being ducked at any point.

When does this not work well?

Some framing cannot survive a circle. If you are at the edge of the frame, or the shot is already tight on your face, scaling down enough to get margin means either a small head in a big circle or a visible crop of the background.

How do you fix a result that came back wrong?

Say which part of you is too close to the edge. Vizard Agent inspects the frame rather than adjusting blindly, so "my chin is tight" and "there is too much space above my head" are both actionable and lead to different adjustments.

How does Vizard Agent compare to doing it yourself?

By hand you mask a circle, drop the video behind it, and it looks right at the size you are working at. Then you watch it on a phone and your chin is touching the rim, so you scale down, and now the background fills a third of the circle.

By hand Vizard Agent
The source Whatever version is open The clean recording
Scaling Until it fits Until there is margin
Checking At working size At full size, at the chin
Bubbles Spaced by eye Timed to spoken lines
Music Added and hoped Measured against the speech

Common questions

Why not use the edited version? Because its titles get cut into fragments by the rim. Vizard Agent starts clean.

How much margin is enough? Enough to see space under the chin. Vizard Agent checks the actual frame.

Can the circle be bigger? Yes. Vizard Agent can widen the circle rather than shrinking you.

Will it add ticks and timestamps? If you want them. Vizard Agent keeps them small so they do not crowd the rim.

Can the bubbles match what I say? Yes, and that is the point. Tell Vizard Agent which line each belongs to.

What if I move around in the shot? Say so. Vizard Agent can follow you rather than using a fixed crop.

Does the audio change? Only if you add music. Vizard Agent measures the levels against your voice.

Can I have a different shape? Yes — rounded square, or a rectangle with a corner radius. Just ask.

Will my background show at the edges? Some of it. Vizard Agent checks what the rim cuts through.

What about the ending? Vizard Agent can give it a soft tail rather than stopping on a word.

Does it work in a horizontal frame? Yes, though the circle is smaller relative to the frame.

Can I get a version without the bubbles? Yes. Ask Vizard Agent for both.

How many passes does the circle take? Usually two or three, because each one is judged on a real frame.

Can I reuse the look on other clips? Yes. Ask Vizard Agent to keep the treatment for the next one.