How to label a video clearly as AI generated

Last updated 2026-09-04 · 7 min read

What is the short version?

If your video is generated and it looks real, somebody needs to be told. The obligation is yours to know — platforms, regulators and clients all have their own rules — but getting the words on screen correctly is a production job with a specific failure mode.

1. Go to Vizard Agent with the cut and the exact wording.
2. Say where the disclosure sits and how long it holds.
3. Ask it to verify the text on the finished render, not the plan.

What do you need before you start?

The wording, in the form you are required to use it. Vizard Agent will place and check whatever you give it, but it cannot tell you what your obligation is — that comes from the platform, the regulator or the client, and it varies.

• The exact text. Copied, not paraphrased.
• Where it goes. A corner, a band, a full card.
• How long. Throughout, or a stated duration.
• The language. More than one, if you need them.
• The platform. Some have their own labelling as well.

What do you type into Vizard Agent?

Give the words verbatim and ask for a check. The wording matters more than the design here — a disclosure that has been tidied up into something more elegant may no longer be the sentence you were required to show.

Prompt

Variants worth knowing:

• "This exact wording." Stops it being improved.
• "Inside the frame." The failure this page is about.
• "Show me on the final render." Not on the plan.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real generated scene that had to be delivered with a disclaimer on it. The interesting part is right at the end: Vizard Agent burned the text in, looked at the result on a frame, and found that it did not fit.

1. Builds the scene — references, shots, voice and music.
2. Reviews every generated shot on frames before assembling.
3. Assembles the cut and checks the mix.
4. Uploads for review and upscales the master.
5. Checks one insert for drift across its duration.
6. Locates the typeface file for the disclosure text.
7. Burns the disclaimer and finishes the master.
8. Verifies the render and watches the final frames.
9. Re-renders with the corrected disclaimer.
10. Checks the disclaimer text on a real frame.
11. Checks that the text fits the frame.
12. Confirms it sits inside the frame before delivery.

Steps nine to twelve are the whole point. The first burn was wrong — that is the normal case, not a failure — and it was caught because the text was inspected on a rendered frame rather than approved as an instruction.

Step eleven deserves saying plainly: disclosure text is usually longer than a caption and usually placed at an edge, which is exactly where a platform crop or a safe-area margin will take the end of it off. A sentence that reads "This video contains AI generated conten" has failed at its only job.

What does the result look like?

A master with the disclosure legible at the size and position you asked for, present for the full duration you specified, and confirmed by Vizard Agent on frames taken from the delivered file itself rather than from a preview or a plan.

Nothing about the wording was improved. What you supplied is what is on screen, which is the point when the sentence is one you have been told to use.

When does this not work well?

This page is about getting the text onto the screen correctly and nothing else. Neither it nor Vizard Agent can tell you what your disclosure should say or whether you need one at all, and treating either as though it could is the main way this goes wrong for people.

• Not knowing the requirement. That comes from your platform or regulator.
• Wording that is too long. It shrinks below readable.
• Busy backgrounds. A plain band may be needed behind it.
• Platform-specific labels. Some also require a metadata flag or a toggle.
• Several jurisdictions. They may not want the same sentence.

How do you fix a result that came back wrong?

Say what you can actually see on screen. Vizard Agent keeps your exact text, the placement it used and the frame measurements it took, so a clipped or unreadable disclosure becomes a re-render of the overlay layer rather than of the whole video.

• "The end is cut off." Re-measured and re-laid inside the frame.
• "It is unreadable on a phone." Enlarged, or a band added behind it.
• "It disappears too early." Extended to the stated duration.
• "The wording changed." Restored verbatim from what you supplied.

How does Vizard Agent compare to doing it yourself?

By hand this is a title layer, which is trivial, and a check, which is the part that gets skipped. The disclosure is usually the last thing added, after the point where anyone is still watching the whole file back.

Common questions

Do I need a disclosure? That depends on your platform, your client and your jurisdiction. Vizard Agent cannot answer it for you.

Will it write the wording? Vizard Agent can draft one, but use the wording you are required to use.

Where should it sit? Somewhere no platform crop reaches. Vizard Agent checks that for you.

How long should it hold? As long as your requirement says. Vizard Agent holds it for whatever you name.

Can it be in two languages? Yes. Vizard Agent lays both and checks both fit.

Does it need a background band? Over busy footage, usually. Vizard Agent will show you the readability.

Will it survive a crop to another ratio? Only if it is placed for that. Say which ratios you will deliver.

Can it add a metadata flag too? Tell Vizard Agent. On-screen text and file metadata are separate things.

What if the wording is very long? Vizard Agent tells you when it will not stay readable at that size.

Can I put it on a card at the end? Yes. Vizard Agent will, though a persistent label is harder to miss.

Does it check the delivered file? Yes — that is the difference from adding a title layer and hoping.

What about partly generated videos? Same method for Vizard Agent. What the label says is your call.

Can it label each generated shot? Yes, if you want per-shot labelling rather than one for the video.

Why check on the render? Because the burn is where it goes wrong, and a plan that says "bottom left, 24pt" tells you nothing about whether the last word is still there.