8 finished animated ads, one dog problem each, 9 min 18 s in total. Each one opens on a real owner's pain, shows a trainer fixing it at home, and ends on a real review and Book Now. 9:16 for Reels and Stories.
Every knock and delivery sets Bones off, so last Halloween the owner turned the porch light off by eight. Yelling only sounds like more barking to him. The trainer starts with soft knocks and rewards him for staying on his bed. Timed for October, when the doorbell never stops.
Proof in the ad: Richard on Trustpilot
Daisy hides behind the couch and shakes whenever Dad visits. Carrying her over to him makes it worse, and the owner has stopped having people over. The trainer has Dad sit on the floor and toss treats, so Daisy decides when to come over.
Proof in the ad: Gaylene on Trustpilot
An 11-week-old puppy leaves the owner's arms covered in scrapes, and the kids are scared to play with her. Pulling your hand away turns it into a game. The trainer spots that it's worst when the puppy skips her nap, and has the owner swap in a chew.
Proof in the ad: Victoria on Trustpilot
Pepper pees on the rug ten seconds after a 20-minute walk, and the owner dreads coming downstairs. Yelling only teaches her to go where nobody can see. The trainer shows the sniff-and-circle signs and sets a schedule, so the owner is already outside when she needs to go.
Proof in the ad: Igor on Trustpilot
Moose rears up and barks at dogs across the street, so the owner walks him at 6 a.m. to avoid everyone. Yanking him back tells him the other dog is a threat. The trainer comes on the walk and calls out the moment he stiffens, so the owner can turn him away in time.
Proof in the ad: Christel on Trustpilot
Biscuit looks right at the owner at the park and keeps running, so now he's on the leash everywhere. Repeating the call teaches him the first few don't count. The trainer has the owner call once, in a normal voice, and make coming back worth it.
Proof in the ad: A Dog Academy owner review
Waffles jumps on everyone who walks in, and the louder the owner yells “off”, the higher he goes. A visit from the sister ends with her pie on the floor. The trainer rings the bell again and again and rewards Waffles only when he's on his mat.
Proof in the ad: Roland on Trustpilot
Told by the dog: Gus pulls, Mom follows, and her shoulder pays for it. Last week he nearly pulled her off the curb, and a new harness didn't help. The trainer watches one real walk and shows her to stop dead the second the leash goes tight.
Proof in the ad: 80,000+ owners served
Each ad is about 70 seconds long and built from 13 to 18 separate animated shots. Every shot is generated by AI, and the AI does not remember anything between shots. It doesn't know the dog in shot 4 is the same dog as in shot 3, and it doesn't know what the narrator is saying. Keeping the story consistent is a manual job that we repeat for every shot.
First we write a "style lock": one paragraph that every image request repeats word for word, so the felt, wood or clay world never drifts. Then we make a character sheet, which shows the dog and every person from several angles. Every later image is made with that sheet attached as a reference.
A frame from a hand-made stop-motion film of painted wooden toys on a miniature tabletop set. Peg-doll people with round wooden heads and simple painted faces (two small solid black painted dot eyes, a tiny painted smile line, NO eyebrows). Unmistakably wooden toys; nothing photoreal, no plastic, no felt, no clay. No text, letters, numbers or logos anywhere.
Why: the AI otherwise slides toward plastic, clay or photo-real from one shot to the next. "NO eyebrows" is there because the video model likes to add eyebrows to a cartoon dog when it acts out an emotion. Viewers read that as a different dog.
BISCUIT: wooden golden-retriever puppy toy, honey-gold matte paint, darker caramel floppy ears, cream muzzle tip, round black nose, two small black dot eyes, NO eyebrows, sky-blue collar with a small round silver tag, exactly four legs.
Why: every detail a viewer could use to spot a "different dog" is pinned down: coat colour, ear colour, collar, tag and number of legs.
We split the script into clauses and give each one its own shot. If the narrator says "you yelp and pull your hand away", the picture has to show exactly that, at that moment. The first frame of each shot is generated as a still image.
THE OWNER (NO eyebrows, plain dot eyes; sleeves pushed up to the elbows, forearms marked with small red scratch lines) sits on the rug and has just snatched his right hand up and away from BISCUIT, the hand OPEN with fingers spread, shaking it as if it stings; BISCUIT, in side profile on the rug, looks up at the raised hand.
FRAME: vertical 9:16, the FIRST FRAME of an animated shot. The top 28% of the frame is calm wall, because captions sit there.
Why: "OPEN with fingers spread" is there because the first version came back as a clenched fist, which reads as about to hit the dog. "Side profile" keeps the dog's face small and steady for the animation step. The empty top band leaves room for the captions.
Animation costs credits, so every still is reviewed first. Two separate AI reviewers each check it twice, comparing it against the character sheets and against the shot before it:
On the puppy-biting ad this caught 9 problems in 18 images before any animation was paid for. They included a smiling face on the "regret" shot, a puppy painted the wrong colour, a daylight shot dropped into an evening scene, and a bite that didn't show a bite.
Each approved still goes to the video model (Kling 3.0 Pro) with a short motion instruction: one action and one camera move. Long instructions make it worse.
The man shakes his stinging hand in the air while the puppy watches it. Painted faces stay the same. Static camera.
Why: one action keeps the motion believable. "Painted faces stay the same" fights the model's habit of redrawing faces. "Static camera" stops it inventing camera moves that break continuity with the next shot. The clip length is set to the length of the narration it plays under.
Every clip is broken into a grid of frames, 8 per second, and reviewed for the problems video models are known for. These are real problems we caught on these six ads:
Any clip with a problem a viewer would notice is fixed. Usually we change the staging and generate it again, for example showing the dog from the side, or from a higher angle so the pen rails don't hide her. We never cover a problem with a frozen still image.
Even with all the checking up front, a quarter to over half of each ad's animation budget has gone to fixing shots that came back wrong.