The film was entirely real. He travelled to the place, filmed there for four days, and met its people.
But the thumbnail did not satisfy him: the sky was grey in all his footage. So he generated an image of the place with a golden sky and perfect light, and used it.
Clicks went up. And in the first hour a comment arrived from somebody who lives there: "that is not a picture of the place; that lake does not exist." And eighty likes underneath it.
The film was honest, and the thumbnail made it look dishonest. That costs more than a click.
The one rule that settles every case
Before any technical or legal detail, write this sentence down:
Never generate an image the viewer will read as something that actually happened.
That rule alone settles ninety per cent of decisions. A viewer does not object to a generated image they know is an illustration; they object to one presented to them as reality.
Three completely different uses
Confusing them causes most of the trouble:
1. Illustrating a concept. A simplified map, a drawing of an abstract idea, a rendering of something that cannot be filmed such as the inside of a deep well or a machine. A sound and useful use, which the viewer reads as a drawing.
2. A background or design element. A gradient, a texture, an abstract shape behind text. No risk, because nobody reads it as reality.
3. Standing in for reality. A place, a person, an event, a moment. This is where the damage starts, and where you say no.
The thumbnail: a promise, not decoration
A thumbnail is not a pretty cover but a promise: this is the moment you will find in the video. And any thumbnail promising something not in the video hurts twice: the viewer leaves after thirty seconds, and the platform sees that people do not stay.
And what actually makes somebody click is not a perfect image but clarity, contrast and a face with an expression, all of which exist in your own footage, set out in the thumbnail: what actually makes someone click.
Which is why the thumbnail rule is stricter than any other: use a frame from your video, a photograph you took, or a design that looks like a design.
Documentary and travel: not at all
If your work is documentary or travel, your credibility is your entire capital. And one generated image of a place or a person puts everything you filmed under a question: "and this, is it real?"
Do not use generated images even where it looks harmless: a reconstruction of an old moment, a picture of a place you could not reach, the face of somebody who declined to appear. Each of those has an honest alternative: a drawing that looks like a drawing, on-screen text, a silhouette, or a voice with no picture.
Disclosure: when it is required and how
In the first two uses, disclosure is good practice rather than always required, and a line in the description is enough. In any place that could be read as reality, disclosure is required, and it belongs inside the image rather than in the description.
And note one practical point: most platforms now offer an option to flag synthetically made or altered content, and some add a label automatically. So use the option yourself, because a label you place looks like transparency and a label added for you looks like exposure. How to say this to your audience without the subject becoming the whole conversation is in what do you tell your audience about your use of AI.
Rights: three questions before you use one
1. What do the tool's terms say about your rights in the output? Some tools grant you broad rights, some restrict them by account type or forbid commercial use on the free plan. Read that before you build an identity on an output.
2. Does the output resemble an existing work? An image that clearly resembles a known artist's style, or contains a known character, or carries a trademark, puts you somewhere you do not want to be.
3. Does it contain a face resembling a real person? A generated face that looks like an existing human being is a different kind of problem, and you do not use it.
And keep a line in your asset log for every image you use: the tool, the date, and the prompt that produced it. That is the same log where you keep font and music licences.
What gives a generated image away
Your audience spots them more than you think. What gives them away most:
- Text inside the image written in malformed letters, in Arabic especially.
- Hands and fingers in odd numbers.
- Reflections that do not match the light source.
- Architectural detail repeating with no logic.
- Faces smooth in an unnatural way.
- And in thumbnails: lighting so perfect it exists in no camera.
And if you need to inspect an image before publishing, magnify its edges, people's hands and any written text, because those are where it shows.
Where they genuinely help
Since this is not a blanket ban, here are places they help with no risk:
- An abstract background for a title or a chapter card.
- A diagram of an idea you are explaining.
- A rendering of something that cannot be filmed: a cutaway of a machine, or a historical scene that looks like a drawing rather than a photograph.
- Visual ideas for yourself: trying a composition or a colour before you film it.
- Treating a photograph you took: removing a distracting object, or extending the frame slightly.
And that last use is the most valuable: the tool improves your image rather than inventing one instead of it. Where AI helps across the rest of your work and where it ruins it is in where AI helps in video making and where it ruins it.
Cheaper and more honest thumbnail alternatives
Before you generate anything, try these four, because most of them give a better result:
- A frame from your video, enlarged and colour-corrected. And shoot stills of your best moments while filming.
- A shot made for the thumbnail at the end of the shoot day, with a clear facial expression.
- A simple two-colour background with large text, which works well for explainer videos.
- An image from a licensed library, clearly a general picture rather than a moment from your film.
A checklist before publishing
- Is there anything generated in the image that reads as real?
- Does the thumbnail promise something that exists in the video?
- Is any Arabic text in it written in correct letters?
- Did you flag the use in the platform's option where required?
- Did you save the tool, the date and the prompt in your log?
In short
A generated image is a design tool, not a tool of testimony. Use it where people read it as a drawing, and leave it where it might be read as reality. And in a thumbnail specifically, the real frame from your video is always stronger, because it promises what you will actually give, and a promise kept is everything a channel is built on.