When I asked the creator behind one of my first pilots what he cared about most, he named two things: which moments get clipped, and the covers. And he said he hated his current Shorts covers. They had been made with an AI image tool: his favourite characters redrawn, glowing, surrounded by speed lines and particle effects that were never in the show. His word for them was “AI slop”.
He wasn’t being picky. He runs an anime fight-analysis channel, and his audience knows those characters frame by frame. They can tell in a split second when a face has been redrawn by a model, and once they notice, the cover stops saying “this is a moment from the video” and starts saying “somebody pushed a button”.
That conversation changed how I make covers. Here’s the approach I use now, and why.
What a Shorts cover actually does
In the Shorts feed the video just starts playing, so the cover isn’t what people see there. The cover matters in the other places: the Shorts shelf on a channel page, search results, and the grid people scroll when they land on your profile after liking one clip.
In those places the cover has one job. It has to tell someone, in less than a second, what this clip is and why it’s worth tapping. That’s a promise about the video. And the fastest way to break that promise is to show something that isn’t in the video.
A real frame is honest by default. What you see on the cover is what you get when you tap.
Why AI-made covers backfire
I understand why people use image generators for covers. They’re fast, they look “designed”, and they fill a blank space with lots of energy. But on Shorts covers they tend to fail in the same few ways:
- Faces drift. Characters and people come out slightly off-model. Fans of a show, and viewers of a person, notice this immediately.
- The effects are fake. Glows, lightning and particle bursts that don’t exist in the footage make the cover look like an ad for a different video.
- Everything looks the same. When a lot of channels use the same tools, the covers start to share one look. Your grid ends up resembling everyone else’s.
- It signals low effort. That’s the part the creator reacted to. A generated cover tells the viewer that nobody spent time on this, and they wonder whether the clip got more care.
None of that is a rule against design. It’s a rule against inventing pictures that aren’t in the video.
How I build a cover from real frames
This is the process I follow now for every batch.
1. Pull frames from the best source you have
Take frames from the highest-resolution original file, not from the finished short. A frame grabbed from a 1080p export looks soft once it fills a whole cover. If the creator has a 4K master, I use that. It costs nothing extra and the difference is obvious on a phone screen.
2. Pick faces and reactions, not wide shots
The frames that work are close-ups with a strong expression: shock, a grin, a stare-down, someone mid-shout. Wide shots and busy frames turn into mush at thumbnail size.
In the anime case, the creator’s own edits sometimes already had speech bubbles or on-screen text in them. Those work well as covers, because the hook is built into the frame.
What I avoid:
- frames in the middle of a glitch or flash transition
- manga panels or frames where the footage is shown inside another frame (like a TV inset)
- anything blurred by motion
3. Choose a layout: one frame or two
There are two layouts I keep coming back to.
The split. Two real frames stacked, one on top and one on the bottom, with a short label on the seam between them. It works well for anything with two sides: a fight, a debate, a before and after, a question and an answer. When I looked at how bigger channels in the same niche packaged their Shorts, one of them used exactly this stacked two-frame style, and another used the single-frame layout below.
The single frame. One clean frame of one character or person, with a short line of text at the top. This works when one face carries the moment, or when the Short’s title already does the explaining.
For the creator above, the split won. When he reviewed our second batch, he told me simpler is better, and that the covers he liked were the two-frame ones with a couple of words on them. He was also fine with a single frame when the title was strong.
4. Keep the label to one or two words per line
The label isn’t a title. It’s a tag that makes the frames make sense. One or two words per line, big, in white or one accent colour, and never across someone’s face.
The words should come from the creator’s own vocabulary. If they say “cooked”, the cover says “cooked”, not “Epic Defeat!”. The two covers he singled out were each just two short words on two lines. That’s about as much text as a cover can carry before it becomes something to read rather than something to see.
5. Match the channel, not a template
Before I make any covers, I look at the creator’s own best-performing Shorts and their long-form thumbnails: fonts, colours, how much text, where the faces sit. The covers should look like they came from the same channel. The same idea applies to the layout of the short itself: keep the creator’s signature visible instead of forcing a generic style on it.
Getting the cover onto YouTube
One practical note, because it trips people up. YouTube has changed how Shorts thumbnails work more than once. According to YouTube’s help page at the time of writing, you can upload a custom Shorts thumbnail in YouTube Studio on a computer, your account has to be verified, and the recommended aspect ratio for the uploaded file is 9:16. Some older guides still say you can only pick a frame from the video in the mobile app, so check the current help page before you plan around it.
If you can’t upload a custom image, the real-frame approach still helps. Pick the frame you’d have used for the cover and choose it as the thumbnail frame.
A quick checklist
Before a cover goes out, I check that it:
- uses only frames that are actually in the video
- comes from the sharpest source file available
- shows a face or reaction that reads at thumbnail size
- has at most one or two words per line, in the creator’s own language
- doesn’t cover a face with text
- looks like it belongs next to the creator’s other Shorts
A cover can’t save a weak clip, so the moment still comes first; I wrote about how to pick moments that work on their own. But once you have a good moment, an honest cover built from real frames is a small, cheap way to show viewers the clip is worth their time.
If you’d like to see this on your own footage, I can cut a free short from one of your videos, cover included. You keep it either way.