The Four-Tell Check
When an AI video reads as fake, it is almost never the whole clip that is wrong. It is one specific, nameable tell that your eye catches before your brain can explain why. I call this the Four-Tell Check, and I run it on every clip before I post it: hands and small moving parts, edge warping, lighting or shadow mismatch, and motion smear between frames. Find the one that is actually wrong in your clip, and you can usually fix it without starting over.
- Hands, fingers, and small moving parts - the most obvious tell, and the first thing viewers notice
- Edge warping - object boundaries that shift or melt where they meet the background
- Lighting and shadow mismatch - light sources that do not agree with each other in the same frame
- Motion smear and morphing - a subject's shape drifting or blending frame to frame during fast movement
Tell 1: Hands, Fingers, and Small Moving Parts
Hands are still the fastest way to spot AI video, because they involve many small joints moving independently and the model has to guess at all of them at once. The fix is not to keep re-rendering the same close-up and hoping. Keep hands out of frame when the shot does not need them, favor wider shots where fingers are not the focal point, and write prompts that specify simple, still poses instead of fidgety or gesture-heavy hand movement.
Tell 2: Edge Warping and Object Boundaries
Edge warping shows up where one object meets another, a hand against a jacket, a coffee cup against a table, hair against a shoulder. The boundary shifts or melts slightly as the clip plays, which reads as wrong even to someone who could not tell you exactly why. Cluttered scenes with many small objects near the frame edges make this worse, because the model has more boundaries to get right at once. Clean, uncluttered backgrounds with fewer overlapping objects near the edges of the frame reduce how many boundaries can go wrong.
Tell 3: Lighting and Shadow Mismatch
A shadow falling the wrong direction relative to the light source in the same frame is one of the fastest tells for anyone who has spent time around a camera, even if they could not name what felt off. This usually comes from a prompt that leaves lighting vague or implies more than one light source without saying so. Name a single, specific light source and time of day directly in the prompt, late afternoon sun through a window, one overhead lamp, so the model has one consistent thing to render shadows from instead of guessing at several.
Tell 4: Motion Smear and Morphing Between Frames
Morphing is when a subject's shape drifts or blends into something slightly different as it moves, most visible during fast or complex motion. Simple camera moves, a slow push in, a steady pan, hold up far better than a shot asking for rapid multi-subject motion all at once. If a scene genuinely needs a complex action, break it into two shorter generations and cut between them rather than asking one continuous clip to carry the whole motion. This is also where routing the shot to the right production model matters, since Seedance 2.0, Kling 3.0, Nano Banana Pro, and Cinema Studio each handle motion differently, and model-routed prompts exist specifically so you are not guessing which one fits a given shot.
Run the Check Before You Post, Not After
The habit that actually saves time is running the Four-Tell Check on a clip before it goes anywhere, not after a comment section points it out. Watch the clip once for hands, once for edges, once for lighting, once for motion, rather than trying to catch all four in a single pass. Most fixes happen in the prompt, not in editing after the fact, which is the whole reason model-routed prompts and a clear single light source description matter more than any post-production trick.
Short, practical drops on income paths, prompts that work, packaging for views, and getting paid. No spam, unsubscribe anytime.
Frequently asked questions
Why do AI videos look fake even when the scene itself looks realistic?
It is almost always one specific tell, not the whole scene. Hands and small moving parts, edge warping between objects, a lighting or shadow mismatch, or motion smear during fast movement are the four most common culprits. Fixing the one that is actually wrong is faster than re-rendering the whole clip.
Can better prompting actually fix these issues, or do I need a different tool?
Prompting fixes most of them. Naming a single consistent light source, simplifying camera movement, keeping cluttered small objects away from frame edges, and routing a shot to the production model built for that kind of motion solve the majority of these tells before you ever generate.
Which tell is the most common one to check first?
Hands and small moving parts, because they involve the most independent joints and are the first thing a viewer's eye catches. Check that one first, then move through edges, lighting, and motion.
Does the production model I choose actually affect how fake a video looks?
Yes. Different production models handle motion and physics differently, so a shot with fast or complex movement can look noticeably better on one model than another. Routing shots to the model actually built for that kind of motion is part of the fix, not just the prompt wording.
Should I fix these issues in editing after generating, or before?
Before, at the prompt level, whenever possible. Editing after the fact can hide some tells with cuts and pacing, but it does not fix the underlying render. Running the Four-Tell Check before you post catches most problems while they are still cheap to regenerate.
Last reviewed by David on August 11, 2026
