How to Judge AI Video and Spot the Tells

You now know how AI video is made and where it tends to go wrong. This last lesson turns that into a skill you can use. You will learn a simple way to judge the quality of an AI clip, the most common signs that a video was made by AI, and why those signs are not enough on their own.

What You'll Learn

• A five point check for judging any AI video clip
• The typical tells of AI video, and where they come from
• Why sound can give a clip away
• Why visual tells fade over time
• How to check a video you find online, beyond looking at it

Watch with a purpose

Most AI clips look good on a first watch. Your eye follows the main subject and fills in the rest. To judge a clip well, watch it more than once, and look for something different each time.

A useful habit:

1. Watch once at normal speed for the overall feel.
2. Watch again and follow one thing only, like a face or a hand.
3. Pause on a few frames, especially where things touch or move fast.
4. Watch the background, not the subject.

A five point quality check

Each point links back to something you learned in this course.

1. Consistency. Compare the start and end of the clip. Does the face match? Do small details like earrings, logos, and buttons stay put? Does anything change after leaving the frame? (Lesson 2)

2. Motion. Watch the hands and feet. Do fingers stay the same in number and shape? Do feet stay planted or slide? Do legs and arms cross in a natural way? (Lesson 3)

3. Physics. Look at any moment where things touch. Does a cup fall with weight? Does liquid pour down? Does a change, like a bite or a cut, stay? (Lesson 3)

4. Detail. Look at text, signs, and fine patterns. Are they readable and stable, or do they melt and shift? Small, distant faces and text are the first things the latent squeeze loses. (Lesson 1)

5. Prompt match. If you made the clip, does it show what you asked, and do events happen in the order you described? Text is weak at describing motion, so check the action, not only the look. (Lesson 4)

A clip that passes all five is good for its purpose. A clip that fails one point may still be usable if the failure is small or easy to cut out.

The typical tells of AI video

When you are trying to work out if a clip is AI made, the same weak spots become tells. Common ones:

• Morphing. One thing smoothly turns into another, like a bag that becomes a coat.
• Melting text. Signs, labels, and screens with letters that wobble or make no sense.
• Sliding feet and odd steps. People glide, or their legs swap in a strange way.
• Hands that change. Fingers merge, split, or change count.
• Too smooth and too clean. Skin with no pores, light that is too perfect, motion with no small shakes.
• Background that rearranges. Windows, trees, or cars shift when you look closely.
• Short shots. The whole video is made of shots of only a few seconds, often with a cut right before something hard would happen.
• Broken cause and effect. A door opens after the person walks through it, or a spill reverses.

Listen as well as look

If the clip has sound, check it too. Some models make audio along with the picture, and others have sound added later. Signs to listen for:

• Speech that does not quite match the lip movement
• Footsteps or impacts that land a little early or late
• Background sound that does not fit the space, like an echo in a small room
• A voice that sounds flat or slightly off in rhythm

Why tells are not enough

Every tell in this lesson comes from a current weakness. As models improve, the weaknesses shrink. Hands are much better than they were. Text is getting better. Clips are getting longer. A clip with no visible tells may still be AI made.

The reverse is also true. Real footage can be blurry, oddly lit, or badly compressed, which can make it look strange. A tell suggests you should look closer. It does not prove anything by itself.

For video that matters, such as news, a public claim, or something that could harm a person, looking is only the first step. Better checks include:

• Find the source. Who first posted it? Is it a known, trusted account or outlet?
• Look for other angles. Real events are often filmed by more than one person.
• Check the context. Does the time, place, and weather match other reports?
• Look for labels. Some companies add invisible watermarks to AI video, and some platforms show labels or content credentials. A label helps, but a missing label proves nothing.

For a full, practical guide to checking videos and images you find online, including deepfakes, take AI Literacy: Spot AI Content & Misinformation.

Putting the course together

You now have the full picture:

• AI video starts from noise and cleans up the whole clip at once, in a squeezed latent space.
• It has no fixed memory of the world, so details drift.
• It learns how motion looks, not physics, so contact, hands, and cause and effect fail.
• Cost and memory keep single clips short.
• Prompts and images steer each step, and more input means less guessing.

Those five ideas explain almost every strength and weakness you will see in any AI video tool.

Key Takeaways

• Judge a clip by watching it several times, each time following something different.
• Use the five point check: consistency, motion, physics, detail, and prompt match.
• Common tells include morphing, melting text, sliding feet, changing hands, and a too clean look.
• Sound can also give a clip away through poor lip sync or badly timed effects.
• Tells fade as models improve, so never rely on looks alone for video that matters.
• For important clips, check the source, other angles, context, and labels.