← All QuickAds reports
Creative Intelligence · Production · September 2026

AI Ad Creative: what works and what still breaks.

Which shots hold up on camera, which still fail, and the one check that catches each category's worst failure. Every page on this subject is written by someone selling AI creative or selling the alternative. We sell one of them, so these limits are drawn against our own interest.

6 sec
Below which most models will do a camera move or a subject action, not both
Hands
The single most re-run shot in the category
7
Recognisable ways generated video fails, from our own QC taxonomy
23%
Warp our own detector measures on clips that are actually fine
Where the line sits, September 2026
Product silhouette across a cutHolds
Brand colour under one lighting setupHolds
Face, single subject, mid-shotHolds
Packaging front-of-pack, static angleHolds
Talking head, under 15 secondsHolds
Skin texture at conversational distanceHolds
Fingers near a faceBreaks
Diamond fire and true dispersionBreaks
Granular physics, pour, scatter, crumbleBreaks
Full-body locomotionBreaks
Bag and accessory scale against a bodyBreaks
Camera move plus action, under 6 secondsBreaks
Section 01

Who this report is for, and our conflict of interest

Stated up front, because the whole value of this document depends on it.

Read this if you are

  • A head of design deciding which shots you can generate and which still need a camera
  • A CMO being sold "AI does everything now" by one vendor and "AI does nothing" by another
  • A brand or creative reviewer who keeps rejecting generated work and cannot articulate why
  • Anyone about to cancel a shoot on the assumption it can be generated

Our stake, plainly

  • QuickAds sells AI creative production. We are not neutral on this question
  • Every limit below costs us something to admit, which is the only reason to trust the list
  • The checks come from our own QC harness, run on our own output. They are not a competitor audit
  • Our operating line is AI handles form, humans own substance. This report is the form half

Section 02 · Executive summary

Five things to know before you brief

01
The shape of the problem

It is not "is AI 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. Quality is not a property of the tool, it is a property of the shot you asked for, which means the decision belongs at shot level in the storyboard, not at tool level in a procurement meeting.

ActionMark every beat in your next storyboard as generate, shoot, or either. Most boards split roughly 70/20/10, and the 20 is always the same handful of shots.
02
The physical tell

Generation renders appearance, not cause

The recurring failure is an effect with no physical cause: fingers hovering rather than gripping, feet not meeting the floor, a cheese pull that stretches forever because nothing tells it to break. Viewers detect this before they can name it, which is why "it looks fine but feels wrong" is the most common review note in the category.

ActionWrite contact into the prompt rather than outcome. "Fingertips pressing the string onto the fingerboard, string visibly depressed" outperforms "playing beautifully", every time.
03
The hard limit

Under six seconds, a model does the move or the action, not both

Ask for a push-in and a subject doing something in a five-second clip and most models silently pick one. The move becomes a drift, or the action stalls. This is the most useful single constraint we have found, because it is predictable and it is cheap to design around.

ActionIf the camera move carries the shot, make it the only motion in the prompt. If the action carries it, cut the move. If you need both, extend the duration.
04
Category-specific

Every category has one check that kills more work than all others 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 and whether the bullet holds its shape once used. Fashion is scale against a body. None of these are generic quality checks, and none are in any tool's default QA.

ActionWrite your category's one check at the top of every review. Section 05 has ours for six categories.
05
Honesty about the detectors

Automated artifact detection is not reliable enough to block on

Our own warp detector measures up to 23% on clips that are completely fine. Hand and face distortion detectors behave similarly. They are useful as a warning that directs a human to a timestamp; they are not useful as a gate. Anyone selling automated AI-artifact detection as a pass/fail decision is overselling it, and that includes anyone selling ours that way.

ActionUse detectors to route attention, not to approve work. A frame pass with human eyes remains non-optional.

Section 03

What holds

Worth stating first, because the failure list is longer and gives a misleading impression. These are shots we now run without a second thought.

ShotWhy it holdsWatch for
Product silhouette across a cutShape is the strongest signal a model carries between framesDrift on reflective or faceted objects
Brand colour, single lighting setupOne condition means one value to holdColour walks when the setup changes mid-cut
Face, single subject, mid-shotThe most-trained subject in every modelIdentity drift across separately generated clips
Skin at conversational distanceTexture reads correctly until the camera gets closePlastic skin in macro; pores are the tell
Front-of-pack, static angleA flat surface held still is close to an image taskFalls apart the moment the pack is tilted
Talking head under 15 secondsShort duration limits how far anything can driftLip sync and dead eyes
Fabric drape, mid and wideCloth behaviour is well modelled at distanceSeams and stitching in close-up
Environmental b-rollNo product fidelity requirement, no handsSignage and text in the background

The pattern in the left column

Everything that holds is either a single subject, a single lighting condition, a short duration, or a shape rather than a mechanism. Everything in the next section violates at least one of those. That is the whole heuristic, and it predicts new failures better than any list of known ones.


Section 04

What breaks

Seven recognisable classes, taken from the failure taxonomy in our own clip QC. They recur across models and across categories.

Failure classWhat it looks likeCost when missed
Action without physical causeFingers hovering rather than gripping, feet not meeting the floor, contact that never happensViewer feels it before naming it. The ad underperforms and nobody can say why.
Localised render artifactsFused fingers, a colour blob in a hand, melted geometry at a joinBrand review rejection, usually at the last approval stage
Continuity breaksObject count changes between shots, a label rewrites itself, a garment changesIn jewelry and FMCG this makes it a different product, not a different shot
Prop leaves and never returnsThe acting object drifts out of frame mid-clip and the shot stops readingThe beat no longer shows what it was written to show
Prompted direction ignoredA push-in that is really a drift, a slow move that never startsThe board and the cut no longer match. Usually found in edit.
Prompted detail absentThe steam, the condensation, the dust that was the entire texture of the shotTechnically passes every automated check
Composition against placementSubject sitting in the 9:16 caption band, half the frame deadFine in the preview, unusable in the placement
Where the failures cluster, by shot element
Hands, especially near a face
Highest
Granular and fluid physics
High
Full-body locomotion
High
Small faceted or reflective product
Medium
Text and logo lockups on a moving surface
Medium
Single face, mid-shot, static
Low
Ordered by how often a beat gets re-run in our own production, not by a published benchmark. Directional and first-party, stated as such rather than dressed up as research.

Faces are easier than legs

Counter-intuitive and consistent. A face is the most-trained subject in any model; a full-body run across a room is a coordinated mechanical sequence with a dozen joints that all have to agree. If a board calls for someone sprinting, budget re-runs for it and do not budget them for the close-up.


Section 05

The one check per category

Six months of production, reduced to the single check that kills more work than everything else in that category combined.

CategoryThe checkWhy this oneStill shoot
JewelryCount the prongs in every shotA prong that migrates between frames means it is not the same ring, and in jewelry the ring is the entire product. Then the shank, where band meets setting, and that join softens. Then the reflection, which has to agree with the object above it.Diamond fire. True dispersion renders as generic white sparkle. For a solitaire hero at full screen, shoot it.
BeautyDoes the skin around the mouth still read as skinThen: does the case hold the same colour in macro and in wide, does the shade hold under two lighting setups, does the bullet keep its shape once used.Any shot with fingers near the face. It is the most re-run shot we have.
FMCG and packaged goodsWould it pass as the pack you would pick up in a shopNot whether it looks good. A pack has a fixed logo lockup, colour, front photograph, weight and tier. Slightly wrong is not an ad, it is a counterfeit that brand review kills and a shopper does not recognise.Pours, scatters and anything granular. Grains moving as a clump is the giveaway.
FoodChar, basil, and whether the pull breaksReal ovens leave uneven blistering and near-burnt spots; generation smooths to an even golden ring that reads as plastic. Leaves wilt where it is hot and stay flat where it is not. Cheese stretches, thins and breaks. A pull that goes on forever is animation.Full-body movement. Faces are easier than legs.
Fashion and accessoriesScale against the bodyBag and accessory sizing is the hardest solved problem in the category and the one generic tools get most visibly wrong. A bag that reads one size in the wide and another in the close-up is the commonest fashion tell.Seams, stitching and hardware in macro.
Supplements and regulatedIs every word on the label legible and correctA label carrying printed claims, on screen and readable, is both a fidelity problem and a compliance one. Wrong text on a regulated pack is a different category of mistake.Anything where the claim itself is the creative.

Why these are worth more than a generic QA list

Every one is checkable by anyone in ninety seconds, against a clip they did not make. That is deliberate. A limitation you can verify is worth more than a capability you have to take on faith, and it is the only reason to believe the rest of this document.


Section 06

The six-second rule

The most useful constraint we have found, because unlike most limits it is predictable, and designing around it costs nothing.

What a model will execute, by clip duration
Under 6 sec · camera move requested
Move or action
Under 6 sec · move only
Reliable
Under 6 sec · action only
Reliable
Over 6 sec · move and action
Usually holds
Below roughly six seconds most models will not execute a camera move and a subject action. They pick one, silently. The board says push-in; the cut has a drift. Observed consistently across models in our own production rather than published by any vendor.
What this breaks

The shots that quietly fail

01
Push-in on someone using the product
Two motions, five seconds. The move degrades to a drift and nobody notices until the edit.
02
Orbit around a hero product
If anything on the product also moves, one of the two stops
03
Pull-back reveal with action in frame
The classic reveal beat, and the classic failure
What to do instead

Three options, in order of cost

Free
Pick one motion
Decide which carries the beat. Cut the other from the prompt entirely rather than softening it.
Cheap
Extend the duration
Past six seconds both usually survive. Often the cheapest fix available.
Better
Split the beat
Two clips, one motion each, cut together. Reads better than a compromised single take.

Section 07

How to actually check it

Three stages. The first two are arithmetic and can be automated. The third needs a person looking at frames, and it is the one that catches the faults that cost a reshoot.

StageWhat it catchesAutomatable
A · TechnicalResolution downgrades, black or frozen frames, chroma blowout, colour jumps, silent or under-level audioFully
B · BrandPalette conformance, logo presence, safe zones, banned and required wording: the part of a guideline that lives in pixelsFully
C · FramesWhether the shot shows what it was written to show. Contact, continuity, camera move, absent detail, composition against placementNot at all

Stage A catches the expensive boring ones

A model quietly returning 720p when you paid for 1080p is caught at the clip stage rather than discovered three days later. Worth automating for this alone.

Stage C is not skippable

Stage A measures pixels. It has no idea what the shot was for, so it will pass a clip that does not depict the thing it was written to depict. A frame pass where nobody looked at a frame is not a frame pass.

Compare panels, do not scrub

Twelve frames in a grid with timestamps burned in. The question that finds the expensive faults: is anything present early and gone later, or the reverse.

What we will not claim about our own detectors

The AI-artifact detectors (warp, hand instability, face distortion) are uncalibrated. Ours measures warp of up to 23% on clips that are known to be fine. They warn; they do not block, and we do not let them trigger a regeneration on their own. Any vendor presenting automated artifact detection as a clean pass/fail is describing something that does not currently exist.


Section 08

What to change when it fails

A finding is half the job. The change is usually one clause in the prompt, not a new brief.

What went wrongWhat to change
Artifacts in hands or facesReframe away from the problem area, or switch to image-to-video from a still that is already clean. Reduce how much is moving at once.
Action has no physical causeName the contact and its consequence, not the outcome. Then drive from an approved first frame rather than from text.
Prop leaves frame and never returnsLock the opening geometry with a generated still showing correct contact, pass it as the first frame, and add an explicit persistence clause. Consider a shorter duration.
Camera move ignored or reversedDrop the move or extend the duration. If the move matters, make it the only motion in the prompt.
Subject in the caption bandA reframing instruction is usually enough. Changing aspect ratio is the heavier option.
Product colour or label driftingFewer lighting conditions in one cut. Every change of setup is another chance for the SKU to walk.

One change at a time

Rewriting six clauses at once means the next clip tells you nothing about which one mattered. This is the most commonly ignored rule and the most expensive.

Image-to-video is the escape hatch

When a shot keeps failing, generating a clean still first and animating from it solves a surprising share of hand, face and product-fidelity problems in one move.

Never re-roll on a warning alone

Uncalibrated detectors are not grounds to spend money. Look at the frame the warning points to, then decide.


Section 09

Where to spend a shoot day

The useful question is not whether to use AI. It is which four shots on the board still justify a camera, and the answer is consistent enough to plan against.

Generate

Most of the board

Volume
Variants of a proven concept
Twenty versions of a working hook. No shoot economics survive this requirement.
Catalogue
The long tail of SKUs
The 95% of the catalogue that never gets a shoot. This is the strongest case in the category.
Context
Environments and b-roll
No product fidelity requirement, no hands, no physics
Localisation
Same concept, new market
Reshooting per market is where budgets historically disappeared
Shoot

The four that keep earning it

01
The hero on a small faceted product
Solitaire at full screen. Dispersion does not render.
02
Anything granular or fluid
Pours, scatters, crumbles. Physics, not appearance.
03
Hands doing something intricate
Especially near a face. The most re-run shot there is.
04
Full-body movement
Running, dancing, sport. Faces are easier than legs.

What this does to a shoot budget

It does not remove the shoot. It changes what the day is for. A day spent capturing four hero shots that generation cannot do, which then feed a hundred generated variants, is a different proposition to a day spent capturing a hundred shots at a hundred-shot cost. That is the honest version of the argument, and it is more useful than either the replacement claim or the dismissal.


Section 10

The plan

Two moves per phase. Nothing here needs new software.

This week

Mark up the board

Go through your next storyboard and mark every beat generate, shoot, or either. The shoot column is usually four beats and always the same kinds.
Write your category's one check at the top of the review template. Prongs, front-of-pack, char, skin, scale, whichever applies.
On the next batch

Change how you prompt and how you review

Rewrite every action line as contact and consequence rather than outcome. This single change removes a large share of the physical-cause failures.
Review on a twelve-frame contact sheet with timestamps, not by scrubbing. Ask what is present early and gone later.
Standing

Set the rules before you need them

Under six seconds, one motion per clip. Make it a production rule rather than a per-shot decision.
Never regenerate on a detector warning alone. Look at the frame it points to, then decide, then change one thing.

The line will move, and this document will date

Everything here describes September 2026. Hands were meaningfully worse a year ago and will be better a year from now. The durable part is not the list. It is the heuristic underneath it: single subject, single lighting condition, short duration, shape rather than mechanism. Anything violating one of those is where to point the review.

About QuickAds

QuickAds.ai, AI Creative Engine

Research winning ads. Generate high-converting creatives. Ship them at volume. All under one roof. The limits in this report come from six months of solving them, per category, on our own production.

32M+

Ads in Library

35+

Languages

23K+

Brands Trust QuickAds

Ad Intelligence

Search and analyse competitor ads across Meta, Google and TikTok. Filter by industry, format, hook type, engagement, brand and market. The 32M+ library is updated continuously.

Creative Production

Generate high-performing ad creatives using AI: static, video, carousel. 1200+ creatives a month per brand, five to seven days from brief to first full batch. AI handles form; humans own substance.

Want to scale creative production without scaling the team?

Visit quickads.ai →