What this report settles
The question is not whether AI creative is good enough. It is which shot. The same model that renders a flawless mid-shot of a face will fuse the fingers when a hand comes near it, so quality is a property of the shot you asked for rather than of the tool, and the decision belongs in the storyboard rather than a procurement meeting.
Everything on this page is drawn from the report itself. The examples are real and the figures are the ones inside it.
The questions below are answered in full. The rest of the report opens as soon as you tell us who you are.
The checks come from the QuickAds QC harness, run on QuickAds output. This is not a competitor audit, and every limit listed costs us something to admit. Every number on this page is an observed benchmark from the ad set described above. None of it is a projection, and none of it is a performance QuickAds is promising your account.
Six questions this report settles with a number
Each answer is quoted from the study and names the ad set it was observed in.
What can AI video still not do well?
Question 01 of sixHands near a face, granular physics like a pour or a scatter, full-body locomotion, diamond fire, and a camera move plus a subject action inside six seconds. Twelve shots are ruled on individually, because the failure is a property of the shot rather than the model. From the QuickAds report on AI creative limits.
Is AI good enough for advertising yet?
Question 02 of sixThat is the wrong question. The same model that renders a flawless mid-shot of a face will fuse the fingers when a hand comes near it, so the decision belongs at shot level in the storyboard, not at tool level in a procurement meeting. From the QuickAds report on AI creative limits.
Why does AI video look fine but feel wrong?
Question 03 of sixBecause generation renders appearance without cause. Fingers hover rather than grip, feet do not meet the floor, a cheese pull stretches forever because nothing tells it to break. Viewers detect it before they can name it. From the QuickAds report on AI creative limits.
Why do AI clips lose the camera move or the action?
Question 04 of sixUnder six seconds, most models do one or the other and silently drop the second. The move becomes a drift, or the action stalls. If the camera move carries the shot, make it the only motion in the prompt. From the QuickAds report on AI creative limits.
What should you check before approving an AI-generated ad?
Question 05 of sixYour category's one check, which kills more work than every generic quality note combined. Jewelry is prong count, FMCG is front-of-pack under four lighting conditions, food is char and the cheese break, beauty is skin around the mouth, fashion is scale against a body. From the QuickAds report on AI creative limits.
Can you rely on AI artifact detectors?
Question 06 of sixNot as a gate. Our own warp detector measures up to 23% on clips that are completely fine, and hand and face distortion detectors behave the same way. They route attention to a timestamp; they do not approve work. From the QuickAds report on AI creative limits.
One exhibit, partly open
Six of twelve rulings. The full table gives the reason each one holds or breaks, and what to watch for.
Where the line sits, September 2026
Live rows| Shot | Verdict |
|---|---|
| Product silhouette across a cut | Holds |
| Face, single subject, mid-shot | Holds |
| Talking head, under 15 seconds | Holds |
| Fingers near a face | Breaks |
| Granular physics: pour, scatter, crumble | Breaks |
| Camera move plus action, under 6 seconds | Breaks |
Quality is not a property of the tool. It is a property of the shot you asked for.
Five moves, in order
Drawn from the report. The reasoning behind each one sits in the full study.
- Mark every storyboard beat as generate, shoot or either before production starts.
- Write contact into the prompt rather than outcome. Describe the grip, not the performance.
- Pick one motion under six seconds. Move or action, never both.
- Put your category's killer check at the top of every review.
- Use detectors to direct a human to a timestamp, never to pass or fail a clip.
Built for the people briefing the work
- Heads of design deciding what still needs a camera
- CMOs being sold both extremes by different vendors
- Brand reviewers who keep rejecting generated work
- Anyone about to cancel a shoot assuming it can be generated
Read ad by ad, not modelled
- Checks drawn from the QuickAds QC harness, run against QuickAds' own output.
- Not a competitor audit. Every limit listed costs us something to admit.
- Rulings are dated September 2026 because the models move, and the report says so.
- Detector accuracy is reported including its false-positive rate, not only its hit rate.
Open the full study
Submit the form to download the printable PDF text and data edition, then open the complete online report.
Report unlocked
Your request is saved. The PDF download will start, then the online report will open. If your browser pauses the download, use Download PDF again.
Read it now