Ten ads built on your custom training plan

Ten times. You've said "sit" ten times, a little louder each time, and he's still nose-down in the grass like you're background noise.

A Dog Academy trainer comes to your home and watches the two of you. Most of the time the fix is in how the cue is given, usually when you say it and what your hands are doing. You leave the lesson knowing exactly what to change.

Get your custom training plan today, built around your dog and the cues he keeps missing. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

Over 80,000 owners and pets served. ★ 4.4 on Trustpilot.

Headline: Get Heard the First Time · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt r2 · reference: 2174803016641101.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/2174803016641101.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
RETRY: the last attempt came back 1072x1467 (ratio 0.73, too tall) and its sticker words were too small to read on a phone. Return exactly 1122x1402, and make the coral sticker bigger, about 24% of the frame width, so "GET YOUR CUSTOM TRAINING PLAN TODAY" reads at a glance.
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "YOU'RE ON YOUR TENTH 'SIT.'" / "YOUR DOG HASN'T EVEN" / "LOOKED UP." / "Your trainer shows you how to get his attention first." / "GET YOUR" / "CUSTOM" / "TRAINING PLAN" / "TODAY" / "built for your dog" / "Over 80,000 owners and pets served".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the three hook lines are the biggest type, capitals about 64 pixels. Sticker words about 30 pixels. The resolution line about 28 pixels, the served line about 22 pixels.

## CONCEPT
SOURCE 2174803016641101 is the layout to inherit: a bright photograph fills the top 62% of the frame; a deep navy band fills the bottom 38% and holds a three-line heavy white question, left-aligned; a coral round sticker, tilted about 12 degrees, sits over the lower right of the photo just above the band. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Photograph, 0% to 62%: a sunny suburban backyard, deep blue sky with a few clouds. In the sharp foreground a young black-and-white border collie mix stands nose-down sniffing the lawn, body turned away, one ear flicked back. Behind, soft and out of focus, a woman in jeans holds one hand raised in a "sit" signal. Shallow depth of field, warm late-morning light.
- Coral #F0605A round sticker, 55% to 73% of the height at the right edge, tilted, a little larger than the source's: "GET YOUR" / "CUSTOM" / "TRAINING PLAN" / "TODAY" in bold white on four short lines, "built for your dog" small under them.
- Navy #16283F band, 62% to 100%: at 66% to 85% the three hook lines in heavy white sans caps, flush left at 6% of the width; at 88% "Your trainer shows you how to get his attention first." in light blue #A9C4E8; at 94% "Over 80,000 owners and pets served" small in white.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt r2 · reference: 2174803016641101.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/2174803016641101.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
RETRY: the last attempt's coral sticker was too small to read on a phone. Make the sticker about 34% of the frame width so "GET YOUR CUSTOM TRAINING PLAN TODAY" reads at a glance; everything else stays as described.
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "YOU'RE ON YOUR TENTH 'SIT.'" / "YOUR DOG HASN'T EVEN" / "LOOKED UP." / "Your trainer shows you how to get his attention first." / "GET YOUR" / "CUSTOM" / "TRAINING PLAN" / "TODAY" / "built for your dog" / "Over 80,000 owners and pets served".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the hook lines are the biggest type, capitals about 62 pixels. Sticker words 28, resolution 28, served line 22 pixels.

## CONCEPT
SOURCE 2174803016641101 is the layout to inherit: a bright photograph over the top of the frame, a deep navy band below it holding a heavy white question, a coral round sticker tilted over the lower edge of the photo. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Photograph, 0% to 50%: a sunny backyard under a deep blue sky; sharp in the foreground a young black-and-white border collie mix nose-down in the grass, body turned away; soft behind, a woman in jeans with one hand raised. The top 14% is sky only.
- Coral #F0605A round sticker, 36% to 49%, right side, tilted: "GET YOUR" / "CUSTOM" / "TRAINING PLAN" / "TODAY" in bold white, "built for your dog" small under them.
- Navy #16283F band, 50% to 100%: 53% to 67% the three hook lines in heavy white caps, flush left; 69% to 72% the resolution line in light blue #A9C4E8; 74% to 76% "Over 80,000 owners and pets served" small in white; 78% to 100% plain navy with a faint paw-print texture, no words.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

You've got the little half-jog down. Arm straight out, smiling at the neighbor like this was the plan, while he hauls you toward the corner.

A Dog Academy trainer walks your actual route with you and your dog: the driveway, the corner, the spot where he always lunges for the grass. You learn the handling that takes the pull out of the leash, on the street where it happens.

Get your custom training plan today, built around your dog and your walking route. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

Over 80,000 owners and pets served. “Cody has been like a different dog on walks this week.” Yelp review

Headline: Walk Him. Don't Chase Him. · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt a1 · reference: 1007941515550296.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1007941515550296.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A typographic ad: crisp vector type on a flat colour ground. Any drawing is a clean flat illustration with correct dog anatomy (four legs, one tail).

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "Do you pretend you're" / "jogging" / "when your dog's actually" / "dragging" / "you?" / "Are you" / "tired" / "of being the neighborhood's daily" / "dog-walking spectacle?" / "A plan built for your dog, coached on your street." / "80,000+ owners and pets served" / "GET YOUR CUSTOM TRAINING PLAN TODAY".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the question stack is the biggest type, letters about 70 pixels tall (x-height about 40). The resolution line about 34 pixels, the served line about 28 pixels, the strip words about 38 pixels.

## CONCEPT
SOURCE 1007941515550296 is the layout to inherit: an all-type ad on a warm cream page; big rounded-sans questions in coral fill the frame, centred, line after line; several key words sit inside flat pastel highlighter boxes (pink, mint, lilac, yellow) with the word in white or coral; a reversed coral strip with white capitals runs across the bottom. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Cream #FBF3E6 ground. From 4% to 66% of the height, centred rounded sans in coral #E8594A, reading as two questions: "Do you pretend you're" ["jogging" in a mint box] "when your dog's actually" ["dragging" in a lilac box] "you?" then "Are you" ["tired" in a butter-yellow box] "of being the neighborhood's daily" ["dog-walking spectacle?" in a pink box]. Words inside boxes are white.
- 70% to 74%: "A plan built for your dog, coached on your street." in deep charcoal, centred.
- At the very top, 3% to 7%: "80,000+ owners and pets served" inside its own pale butter-yellow highlighter box (the same flat box style as the question words), charcoal, centred; the questions start below it at 10%. Nothing between 76% and 86% except empty cream.
- 87% to 95%: a coral strip, inset from the edges like the source, with "GET YOUR CUSTOM TRAINING PLAN TODAY" in white capitals.
- A tiny flat line drawing of a dog pulling a person on a leash may sit at the right end of the last question line, no bigger than two lines of type.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt a1 · reference: 1007941515550296.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1007941515550296.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A typographic ad: crisp vector type on a flat colour ground. Any drawing is a clean flat illustration with correct dog anatomy (four legs, one tail).

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "Do you pretend you're" / "jogging" / "when your dog's actually" / "dragging" / "you?" / "Are you" / "tired" / "of being the neighborhood's daily" / "dog-walking spectacle?" / "A plan built for your dog, coached on your street." / "80,000+ owners and pets served" / "GET YOUR CUSTOM TRAINING PLAN TODAY".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the question stack is the biggest type, letters about 64 pixels tall. Resolution 32, served box 26, strip 34 pixels.

## CONCEPT
SOURCE 1007941515550296 is the layout to inherit: an all-type ad on a warm cream page, big rounded-sans questions in coral, key words in flat pastel highlighter boxes, a reversed coral strip. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Cream #FBF3E6 ground. 0% to 14%: a small flat line drawing of a dog pulling a person on a leash, centred, no words.
- 15% to 58%: the two questions centred in coral #E8594A rounded sans with the same four highlighter boxes as the 4:5 (mint "jogging", lilac "dragging", butter-yellow "tired", pink "dog-walking spectacle?"), words in boxes white. The questions run from 20% to 60%.
- 15% to 18%: "80,000+ owners and pets served" in charcoal inside its own pale butter-yellow highlighter box, centred, above the questions.
- 63% to 66%: the resolution line in charcoal.
- 70% to 75%: the coral strip, inset, with the offer in white caps.
- 78% to 100%: cream, empty.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

The doorbell goes and you're already moving: grab the collar, shove him in the laundry room, shut the door, smile. Then you spend the evening listening to him whine through it.

A Dog Academy trainer comes over and runs the arrival with you. Real knocks, the door opening, someone walking in. He learns what to do with four paws on the floor, and you learn how to set it up so dinner guests can meet your dog.

Get your custom training plan today, built around your dog and your front door. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

80,000+ owners and pets served, rated 4.4 on Trustpilot.

Headline: Stop Hiding Your Dog From Guests · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt a1 · reference: 1557838955821570.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1557838955821570.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A polished 3D animated-film still, like a modern studio feature: soft global illumination, stylised but believable characters with correct anatomy, a warm readable scene. The dog has four legs and one tail. No photographic people.

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "ARE YOU HOSTING" / "GUESTS" / "OR HIDING YOUR" / "JUMPING DOG" / "AGAIN?" / "Your trainer rehearses real arrivals at your front door." / "More than 80,000 owners and pets served" / "GET YOUR CUSTOM TRAINING PLAN TODAY".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the headline lines are the biggest type, capitals about 78 pixels. The resolution line about 30 pixels, the served line about 22 pixels, the band words about 36 pixels.

## CONCEPT
SOURCE 1557838955821570 is the layout to inherit: a near-black ground; a huge stacked headline in heavy condensed caps fills the top-left 55%, lines alternating red, white and yellow, with a short hand-drawn red underline; a 3D animated character scene fills the right half and the lower middle; two short lines of small white text sit lower left beside a thin red bar; a red band with a white CTA runs across the bottom. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Near-black #16161A ground with a soft warm spotlight on the scene.
- Top-left, 4% to 44%, heavy condensed caps: "ARE YOU HOSTING" white, "GUESTS" red #E03A2F, "OR HIDING YOUR" white, "JUMPING DOG" yellow #F6C343, "AGAIN?" white, with a short red brush underline under "AGAIN?".
- Right 55%, 18% to 88%: a 3D animated hallway. A sheepish 3D man in a grey sweater leans on a laundry-room door, holding it shut with his back while a big goofy golden retriever's nose and one paw poke out of the gap. Behind him the front door's frosted glass shows two guest silhouettes and a warm porch light. His face says "nothing to see here".
- Lower left, 56% to 68%: a thin red vertical bar and beside it "Your trainer rehearses real arrivals at your front door." in white, then "More than 80,000 owners and pets served" in light grey.
- 89% to 100%: red #E03A2F band, "GET YOUR CUSTOM TRAINING PLAN TODAY" in heavy white caps, centred.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt a1 · reference: 1557838955821570.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1557838955821570.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A polished 3D animated-film still, like a modern studio feature: soft global illumination, stylised but believable characters with correct anatomy, a warm readable scene. The dog has four legs and one tail. No photographic people.

## RULES
LAYOUT: the whole headline block sits from 20% to 44% of the height, over the dark hallway wall; above it, from 0% to 19%, only the dark ceiling and a glowing pendant lamp. The red band ends by 80%. The dog is entirely behind the laundry-room door: only its nose and one paw show through the gap, and no tail or any other part of a dog appears in the hallway.
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "ARE YOU HOSTING" / "GUESTS" / "OR HIDING YOUR" / "JUMPING DOG" / "AGAIN?" / "Your trainer rehearses real arrivals at your front door." / "More than 80,000 owners and pets served" / "GET YOUR CUSTOM TRAINING PLAN TODAY".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the headline lines are the biggest type, capitals about 74 pixels. Resolution 28, served line 22, band words 32 pixels.

## CONCEPT
SOURCE 1557838955821570 is the layout to inherit: a near-black ground, a huge stacked headline in heavy condensed caps alternating red, white and yellow, a 3D animated character scene, small white text beside a thin red bar, a red CTA band. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Near-black #16161A ground. 0% to 14%: dark ceiling of the hallway, no words.
- 15% to 38%: the five headline lines, left-aligned, same colours as the 4:5, red brush underline under "AGAIN?".
- 38% to 66%: the 3D animated hallway scene, full width: the sheepish man in a grey sweater holding the laundry-room door shut with his back, the golden retriever's nose and paw poking out of the gap, frosted front door with two guest silhouettes behind.
- 67% to 71%: thin red bar and the resolution line in white, then "More than 80,000 owners and pets served" in light grey.
- 72% to 76%: the red band with "GET YOUR CUSTOM TRAINING PLAN TODAY" in white caps.
- 78% to 100%: the hallway floor and a doormat in the dark, no words.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

You check every corner before you turn it and walk at 6 a.m. because fewer people are out. You still end up hauling him back while another owner stares.

Dog Academy trainers work on reactivity in person on your route, and between visits you can get live virtual lessons, so there's a real person to ask before your next walk.

Get your custom training plan today, built around your dog and the walks you dread. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

Over 80,000 owners and pets served. “We now have a game plan that works.” Trustpilot review

Headline: A Plan for Every Walk · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt a1 · reference: 2017605652284853.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/2017605652284853.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "DO YOU SCAN" / "EVERY CORNER" / "BEFORE YOUR DOG SPOTS" / "ANOTHER DOG?" / "Get a walking plan for the encounters you dread." / "LIVE LESSON" / "✓ 80,000+ owners and pets served" / "Get your custom training plan today".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the headline lines are the biggest type, capitals about 72 pixels. The resolution line about 30 pixels. On the phone: the banner about 26 pixels, the chip about 20 pixels, "LIVE LESSON" about 18 pixels.

## CONCEPT
SOURCE 2017605652284853 is the layout to inherit: a deep navy ground; four lines of huge condensed caps top-left, white with the second line in orange; a short paragraph of small white text under them; a modern phone standing upright on the right half playing a video of a friendly man; a photo of a dog lunging on a leash at lower left behind the text; an orange pill button across the bottom. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Deep navy #12213A ground.
- Top-left, 4% to 36%: "DO YOU SCAN" white, "EVERY CORNER" orange #F28C28, "BEFORE YOUR DOG SPOTS" white, "ANOTHER DOG?" white, heavy condensed caps; at 38% to 43% "Get a walking plan for the encounters you dread." in white regular.
- Right 44% of the width, 12% to 90%: a black modern phone, straight on, screen showing a live video call: a friendly woman trainer in her 40s in a plain green fleece, kneeling on a sidewalk and smiling, one hand explaining; a small red dot and "LIVE LESSON" chip in the top corner of the screen; a translucent chip lower on the screen reading "✓ 80,000+ owners and pets served"; a white rounded banner at the bottom of the screen reading "Get your custom training plan today".
- Lower left, 50% to 92%, behind nothing but navy edges: a real photograph, blended into the navy, of a German shepherd on a sidewalk lunging and barking at a passing doodle across the street, the owner's arm pulled taut on the leash.
- No button below the phone; the bottom 8% is navy.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt r3 · reference: 2017605652284853.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/2017605652284853.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
RETRY: the last attempt started "DO YOU SCAN" at 10% of the height, under the phone's profile bar. The headline block now sits from 18% to 36% of the height; above it, from 0% to 17%, only the dark street, trees and a streetlight.
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "DO YOU SCAN" / "EVERY CORNER" / "BEFORE YOUR DOG SPOTS" / "ANOTHER DOG?" / "Get a walking plan for the encounters you dread." / "LIVE LESSON" / "✓ 80,000+ owners and pets served" / "Get your custom training plan today".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the headline lines are the biggest type, capitals about 64 pixels. Resolution 28; on the phone the banner 24, chip 20, LIVE LESSON 18 pixels.

## CONCEPT
SOURCE 2017605652284853 is the layout to inherit: a deep navy ground, a stacked headline in huge condensed caps (white with one orange line), a modern phone playing a trainer video, a photo of a dog lunging on a leash blended into the navy. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Deep navy #12213A ground. 0% to 14%: navy with a faint streetlight glow, no words.
- 15% to 33%: the four headline lines, left-aligned, "EVERY CORNER" in orange #F28C28; 34% to 37%: the resolution line in white.
- 38% to 64%: the black phone, centred and upright, a little smaller, showing the live video lesson: the friendly woman trainer in a green fleece kneeling on a sidewalk; "LIVE LESSON" chip with a red dot at the top of the screen; a translucent chip lower on the screen reading "✓ 80,000+ owners and pets served"; the white banner "Get your custom training plan today" at the bottom of the screen.
- 60% to 100%, behind and beside the phone, blended into navy: the German shepherd lunging on a sidewalk at a passing doodle, the owner's arm pulled taut. Below 66% only this photo, no words.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Ten minutes in the yard. Nothing. You come back inside, take off your shoes, and there he is next to a fresh puddle on the good rug.

Accidents right after a walk usually come down to timing and routine. A Dog Academy trainer comes to your home, looks at your actual day, and sets up the routine with you.

Get your custom training plan today, built around your puppy and your daily routine. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

Dog Academy has served over 80,000 owners and pets.

Headline: Keep the Rug Dry · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt a1 · reference: 36294478773499333.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/36294478773499333.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "You just took" / "your puppy out." / "Why is your rug wet?" / "leash still on" / "brand new rug" / "Get your custom" / "training plan today!" / "Your trainer fixes the routine behind those accidents." / "served 80,000+ owners & pets".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the three hook lines are the biggest writing, about 60 pixels tall. The two arrow notes about 34 pixels, the plan note about 36 pixels, the resolution line about 30 pixels, the served note about 26 pixels.

## CONCEPT
SOURCE 36294478773499333 is the layout to inherit: a saturated coral ground; one large overhead hero photograph fills the middle of the frame and runs off the bottom edge; white hand-drawn notes in a loose marker hand sit in the upper-left and upper-right corners, each with a curved hand-drawn arrow pointing into the photo. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Coral #EE6A55 ground.
- Hero photograph, 26% to 100% of the height and 8% to 92% of the width, a rounded-corner cut-out as in the source: straight overhead, a thick cream wool rug on a wooden floor; a small fresh wet spot on the rug; sitting right beside it a golden retriever puppy looking up at the camera with guilty eyes, a red leash still clipped to its collar and trailing off the rug; the edge of a back door mat at the top.
- Top-left, 3% to 20%: the hook in big white handwriting on three lines.
- Top-right, 4% to 16%: "Get your custom" / "training plan today!" in white handwriting on two lines, slightly rotated, with a curved arrow.
- Two small white notes with arrows on the photo edges: "leash still on" pointing at the leash; "brand new rug" pointing at the rug.
- At 21% to 24%, under the hook: "Your trainer fixes the routine behind those accidents." in small white handwriting.
- Bottom-right corner over the coral margin: "served 80,000+ owners & pets" in small white handwriting.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt a1 · reference: 36294478773499333.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/36294478773499333.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
LAYOUT: the top 16% of the frame is picture or background only, and the first word starts at 17% of the height or lower, below the phone's profile bar.
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "You just took" / "your puppy out." / "Why is your rug wet?" / "leash still on" / "brand new rug" / "Get your custom" / "training plan today!" / "Your trainer fixes the routine behind those accidents." / "served 80,000+ owners & pets".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the hook lines are the biggest writing, about 58 pixels tall. Notes 32, the plan note 34, resolution 28, served note 24 pixels.

## CONCEPT
SOURCE 36294478773499333 is the layout to inherit: a saturated coral ground, one large overhead hero photograph running off the bottom edge, white hand-drawn notes with curved arrows pointing into it. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Coral #EE6A55 ground. 0% to 14%: coral with a few small hand-drawn white paw prints, no words.
- 15% to 27%: the hook in big white handwriting on three lines, left; "Get your custom" / "training plan today!" rotated at the right, 16% to 22%, with a curved arrow down.
- 28% to 31%: the resolution line in small white handwriting.
- Hero photograph, 34% to 100%, rounded top corners: overhead, the cream wool rug with a small fresh wet spot, the golden retriever puppy sitting beside it looking up guiltily, red leash still clipped on.
- On the photo's upper edge, 36% to 50%: "leash still on" and "brand new rug" in white handwriting with arrows; at 72% to 75% on the coral side margin: "served 80,000+ owners & pets". Below 78% only the photo.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

11:40, the crying starts. 1:15, you're on the floor next to the crate with your fingers through the wire. 3:30, you give up and bring him into bed. 6:30, the alarm.

A Dog Academy trainer comes to your home and sets up the crate, the evening and the last potty trip with you, so the crate becomes somewhere he settles. The plan is built around your house and your schedule.

Get your custom training plan today, built around your puppy and your nights. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

More than 80,000 owners and pets served.

Headline: Sleep Through the Night Again · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt a1 · reference: 1266203765370801.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1266203765370801.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
Clean flat illustration panels in one consistent style, with correct dog anatomy (four legs, one tail); any photo panel is a real, well-lit photograph.

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "Did you want a puppy" / "or another night without sleep?" / "11:40 PM" / "Crying starts" / "1:15 AM" / "Still crying" / "3:30 AM" / "You give up" / "6:30 AM" / "Alarm" / "Your trainer builds a crate routine that lets you sleep." / "80,000+ owners and pets served so far" / "Get your custom training plan today." / "Built for your puppy.".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the two title lines are the biggest type, about 56 pixels. Tile times about 40 pixels, tile captions about 24 pixels, the resolution line about 26 pixels, the offer tile words about 30 pixels.

## CONCEPT
SOURCE 1266203765370801 is the layout to inherit: a charcoal ground; a bold white title across the top; below it a grid of dark rounded tiles with thin gaps — one big hero tile at top-left (about 60% of the width, 55% of the height) with a large photo and a bold title, a column of two smaller tiles on the right, and a row of three small tiles along the bottom; every tile has a bold title and a photo. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Charcoal #1E1F22 ground; tiles #2A2C30 with 16-pixel rounded corners.
- Title, 3% to 14%: "Did you want a puppy" / "or another night without sleep?" bold white, centred.
- Hero tile, 17% to 64%, left 60%: a real photograph in soft blue night light of a black lab puppy sitting in a wire crate with its head up, mid-howl; top-left of the tile "11:40 PM" in warm yellow #F4C542 and "Crying starts" in white.
- Right column: tile one, 17% to 40%: an owner's hand resting on the crate wire in the dark, "1:15 AM" and "Still crying"; tile two, 41% to 64%: the puppy asleep across a pillow in a bed, "3:30 AM" and "You give up".
- 66% to 70%: "Your trainer builds a crate routine that lets you sleep." in white, centred.
- In the hero tile's top-right corner, a small dark translucent chip with "80,000+ owners and pets served so far" in white.
- Bottom row, 72% to 97%, two tiles: left (55% wide), a phone alarm glowing on a nightstand, "6:30 AM" and "Alarm"; right, a warm yellow #F4C542 tile with "Get your custom training plan today." large and "Built for your puppy." under it, in charcoal.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt r2 · reference: 1266203765370801.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1266203765370801.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
Clean flat illustration panels in one consistent style, with correct dog anatomy (four legs, one tail); any photo panel is a real, well-lit photograph.

## RULES
LAYOUT: the top 16% of the frame is picture or background only, and the first word starts at 17% of the height or lower, below the phone's profile bar.
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "Did you want a puppy" / "or another night without sleep?" / "11:40 PM" / "Crying starts" / "1:15 AM" / "Still crying" / "3:30 AM" / "You give up" / "6:30 AM" / "Alarm" / "Your trainer builds a crate routine that lets you sleep." / "80,000+ owners and pets served so far" / "Get your custom training plan today." / "Built for your puppy.".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the two title lines are the biggest type, about 54 pixels. Tile times 38, captions 22, resolution 24, offer tile 28 pixels.

## CONCEPT
SOURCE 1266203765370801 is the layout to inherit: a charcoal ground, a bold white title, a grid of dark rounded tiles with thin gaps, each tile a photo with a bold title. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Charcoal #1E1F22 ground; tiles #2A2C30 with rounded corners. 0% to 14%: charcoal with a thin crescent moon, no words.
- 15% to 23%: the title on two lines, bold white, centred.
- A small dark translucent chip with "80,000+ owners and pets served so far" in white sits in the hero tile's top-right corner.
- 25% to 49%: the big hero tile full width: the black lab puppy mid-howl in the wire crate in blue night light, "11:40 PM" in yellow #F4C542 and "Crying starts" in white.
- 50% to 62%: the wide warm yellow offer tile, full width, with "Get your custom training plan today." / "Built for your puppy." in charcoal.
- 63% to 76%: three tiles: the hand on the crate wire, "1:15 AM" / "Still crying"; the puppy asleep on a pillow in bed, "3:30 AM" / "You give up"; the alarm on a nightstand, "6:30 AM" / "Alarm".
- The resolution line runs in white along the bottom edge inside the hero tile, 45% to 48%.
- 78% to 100%: charcoal, empty.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

The delivery van, the neighbor's cat, a leaf, a car door three houses down. Every evening you say "quiet" forty times and he gets louder.

A Dog Academy trainer comes to your house and works on the barking at the window where it starts. You learn what to do in the first second, before it turns into ten minutes of noise.

Get your custom training plan today, built around your dog and your home. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

Over 80,000 owners and pets served.

Headline: Get Your Evenings Back · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt r2 · reference: 1600917341636107.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1600917341636107.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "How many more evenings will you" / "spend begging your dog for quiet?" / "A trainer works on the barking where it starts." / "Get your custom training plan today." / "Dog Academy · 80,000+ owners and pets served".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the two hook lines are the biggest type, capitals about 50 pixels, bold. The yellow serif line about 40 pixels, the offer line about 26 pixels, the signature about 24 pixels.

## CONCEPT
SOURCE 1600917341636107 is the layout to inherit: one close-up photograph of a dog howling with its snout pointed up, set in a large rounded-corner card with a thin white margin; in the lower left, over the photo's own dark shadow, two lines of bold white sans and under them one line of warm yellow serif; a small signature line at the bottom left. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- White margin around a rounded card that fills 3% to 97% of both dimensions.
- Photograph: dusk in a living room. A tricolour beagle stands with its front paws on the back of a sofa at a big window, head thrown back, mouth open mid-bark, lit by a warm lamp from the side; outside the window, blue dusk and a blurred delivery van. Behind the dog, soft and out of focus, a person on the sofa with both hands pressed over their ears. The lower third falls into deep shadow.
- Top left of the card, 6% to 12%, over the dusk window: "Dog Academy · 80,000+ owners and pets served" small in white, and under it "Get your custom training plan today." in white, like a byline.
- Lower left, 70% to 80%: the two hook lines in bold white sans; 82% to 86%: "A trainer works on the barking where it starts." in warm yellow #F6C343 serif. Nothing else below.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt r4 · reference: 1600917341636107.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1600917341636107.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
RETRY: keep the Dog Academy byline from 20% to 25% of the height, below the phone's profile bar. The last attempt filled the frame edge to edge and read like a different ad in this set; this time, as in the source, the photo is a rounded card inset inside a clean white margin about 6% of the width on every side, so the white frame is clearly visible all around.
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "How many more evenings will you" / "spend begging your dog for quiet?" / "A trainer works on the barking where it starts." / "Get your custom training plan today." / "Dog Academy · 80,000+ owners and pets served".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the two hook lines are the biggest type, capitals about 50 pixels. Serif line 38, offer 24, signature 22 pixels.

## CONCEPT
SOURCE 1600917341636107 is the layout to inherit: one close-up photograph of a dog barking with its snout up, in a large rounded-corner card with a thin white margin; bold white lines and a warm yellow serif line over the photo's shadow; a small signature line. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Rounded card filling the frame inside a thin white margin. Photograph: dusk living room, the tricolour beagle with front paws on the sofa back at the big window, head back mid-bark, warm lamp from the side, blue dusk and a blurred delivery van outside; soft behind, a person on the sofa with hands over their ears. The top 14% is window and dusk sky, no words.
- 16% to 21%, top left over the dusk window: "Dog Academy · 80,000+ owners and pets served" small in white, and under it "Get your custom training plan today." in white, like a byline.
- 58% to 72%, over the photo's shadow, left: the two hook lines in bold white; the resolution in yellow #F6C343 serif. Nothing else below.
- 78% to 100%: the dark sofa and floor, no words.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

He's fine while you're home. The second the door clicks shut, the howling starts. You come back to a scratched door and a note from the neighbor.

A Dog Academy trainer works on leaving with you, in your own home, starting with departures so short he doesn't have time to panic and building from there. You learn what to do before you walk out, and what to do when you walk back in.

Get your custom training plan today, built around your dog and your comings and goings. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

We've served over 80,000 owners and pets.

Headline: Leave the House Without the Howling · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt a1 · reference: 1358665089066870.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1358665089066870.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "80,000+ owners and pets served to date" / "Your dog is quiet beside you." / "Then you leave. Then howling." / "Howling" / "Starts the second the door shuts" / "Scratching" / "The paint on the door is gone" / "Chewing" / "Whatever smells like you" / "Accidents" / "Right by the front door" / "Your trainer practices leaving with you, at home." / "Get your custom training plan today.".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the two title lines are the biggest type, about 54 pixels. Row titles about 34 pixels, sublines about 24 pixels, the resolution line about 30 pixels, the offer line about 24 pixels, the kicker about 20 pixels.

## CONCEPT
SOURCE 1358665089066870 is the layout to inherit: a soft warm light-grey studio backdrop; a centred title in dark grey at the top; four rows stacked in the middle, each with a round outline icon on the left (thin gradient ring), a bold dark-grey row title and a lighter one-line subline; a real photographic scene on a surface across the bottom. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Warm light grey #EDEBE8 ground with a soft window light from the left.
- 3% to 5%: "80,000+ owners and pets served to date" small, centred, light grey. 6% to 16%: the two title lines, centred, dark charcoal.
- 19% to 58%: four rows starting at 22% of the width: a sound-wave icon, "Howling", "Starts the second the door shuts"; a paw-scratch icon, "Scratching", "The paint on the door is gone"; a shoe icon, "Chewing", "Whatever smells like you"; a droplet icon, "Accidents", "Right by the front door".
- 60% to 90%: a real photograph on the same grey, like the source's shelf: the inside of a white front door with scratch marks on the lower edge, a golden doodle lying pressed against the door with its chin on its paws, eyes up, a chewed sneaker beside it on the floor.
- 17% to 19%: the offer line "Get your custom training plan today." centred in a soft teal #2F7F7A, right under the title; the rows start at 22%.
- 92% to 95%: "Your trainer practices leaving with you, at home." centred charcoal. Nothing else below.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt a1 · reference: 1358665089066870.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1358665089066870.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
LAYOUT: the top 16% of the frame is picture or background only, and the first word starts at 17% of the height or lower, below the phone's profile bar.
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "80,000+ owners and pets served to date" / "Your dog is quiet beside you." / "Then you leave. Then howling." / "Howling" / "Starts the second the door shuts" / "Scratching" / "The paint on the door is gone" / "Chewing" / "Whatever smells like you" / "Accidents" / "Right by the front door" / "Your trainer practices leaving with you, at home." / "Get your custom training plan today.".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the two title lines are the biggest type, about 52 pixels. Row titles 32, sublines 22, resolution 28, offer 22, kicker 20 pixels.

## CONCEPT
SOURCE 1358665089066870 is the layout to inherit: a soft warm light-grey studio, a centred title, four rows each with a round outline icon, a bold title and a light subline, a real photographic scene at the bottom. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Warm light grey #EDEBE8 ground. 0% to 14%: plain grey with soft window light, no words.
- 17% to 19%: the kicker, small, centred; 20% to 27%: the two title lines, centred charcoal; 28% to 30%: the offer line "Get your custom training plan today." in soft teal #2F7F7A.
- 32% to 54%: the four icon rows (Howling, Scratching, Chewing, Accidents) with their sublines.
- 54% to 70%: the photograph: the golden doodle lying pressed against the inside of the white front door, chin on paws, scratch marks on the door edge, a chewed sneaker beside it.
- 72% to 75%: the resolution line centred charcoal. Nothing else below it.
- 78% to 100%: the floor and the bottom of the door continue, no words.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

He comes every time in the kitchen. Then the gate's open, there's a squirrel, and you're sprinting across the park yelling his name while he has the best day of his life.

A Dog Academy trainer practices recall with you in the real places he runs, from the kitchen out to the park. You learn how to make "come" worth more than the squirrel.

Get your custom training plan today, built around your dog and the places he runs. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

Over 80,000 owners and pets served. “Now they are pretty calm and I can take them anywhere without worrying.” Trustpilot review

Headline: Stop Chasing Your Dog · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt r2 · reference: 1932982077338407.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1932982077338407.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "Your dog knows 'come.'" / "Why are you still chasing him?" / "Recall practice where he runs, with a trainer beside you." / "Get your custom" / "training plan today." / "Served: 80,000+ owners and pets" / "“Now they are pretty calm and I can take them anywhere without worrying.” Trustpilot review".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the two headline lines are the biggest type, capitals about 64 pixels. The resolution line about 30 pixels, the circled plan words about 46 pixels, the served line about 24 pixels, the review line about 20 pixels.

## CONCEPT
SOURCE 1932982077338407 is the layout to inherit: a full-bleed photograph of bright green grass shot from slightly above, the dog in the middle; a bold white two-line headline centred at the top; a hand-drawn white oval loop around a short white note at the lower left; an object at the lower right. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Photograph: a wide green park on a sunny afternoon, camera low over the grass. In the middle a golden retriever sprints toward the camera's right, ears flying, tongue out, delighted, a long blue training leash trailing across the grass behind it. Far in the background, small and soft, the owner running after it with one arm raised.
- 4% to 17%: the two headline lines, bold white rounded sans, centred, with a soft shadow.
- 18% to 21%: "Recall practice where he runs, with a trainer beside you." in white, centred.
- 22% to 26%: the review line in small white italic, centred, under the resolution line.
- Lower left, 66% to 86%: a big loose hand-drawn white oval around "Get your custom" / "training plan today." in bold white, large, capitals about 46 pixels, and under them inside the oval "Served: 80,000+ owners and pets" smaller. Nothing written below 88%.
- Lower right, 78% to 96%: the end of the blue leash coiled on the grass where the source puts its product. No words on it.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt a1 · reference: 1932982077338407.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1932982077338407.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "Your dog knows 'come.'" / "Why are you still chasing him?" / "Recall practice where he runs, with a trainer beside you." / "Get your custom" / "training plan today." / "Served: 80,000+ owners and pets" / "“Now they are pretty calm and I can take them anywhere without worrying.” Trustpilot review".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the two headline lines are the biggest type, capitals about 62 pixels. Resolution 28, circled note 28, review 20 pixels.

## CONCEPT
SOURCE 1932982077338407 is the layout to inherit: a full-bleed photograph of bright green grass, a bold white two-line headline centred near the top, a hand-drawn white oval around a short note, an object at the lower right. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Tall photograph of a wide green park, camera low: the golden retriever sprinting toward the right in the middle of the frame (40% to 64%), ears flying, a long blue training leash trailing; far behind, the owner small and soft, running with one arm up. The top 14% is sky and treeline, no words.
- 15% to 25%: the two headline lines, bold white rounded sans, centred; 26% to 29%: the resolution line in white; 30% to 34%: the review line in small white italic, centred.
- 64% to 75%: the hand-drawn white oval around "Get your custom" / "training plan today." and, smaller, "Served: 80,000+ owners and pets" at the left. Nothing written below 76%.
- 78% to 100%: grass and the coiled end of the blue leash, no words.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

One minute he's asleep on your lap. The next he's hanging off your sleeve, then your kid's hand, and "no" just makes it a game.

A Dog Academy trainer comes to your home and shows you what to do the moment the teeth come out, during the play that sets it off, with your puppy and your family.

Get your custom training plan today, built around your puppy and your family. It includes up to 8 in-person lessons at your door (1 a week), lifetime unlimited virtual lessons (up to 2 a week) and the full online course library.

Over 80,000 owners and pets served. “My pup went from Cujo to a sweetheart.” Trustpilot review

Headline: Help for Puppy Biting · CTA: Book Now

Feed 4:5 · gpt-5.6-sol · high · attempt r2 · reference: 1249958710576916.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 1122x1402 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1249958710576916.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 4:5 image, exactly 1122x1402 pixels (width 1122, height 1402), nearly square and only slightly taller than wide, not a phone-screen shape. Do not return 1072x1467.
Create ONE complete, finished 4:5 Meta feed ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its grid, type placement, type weights, devices and colour logic, recomposed for 4:5 and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "Your puppy melts" / "your heart. Then" / "sinks teeth into" / "your hand." / "★★★★★" / "“My pup went from Cujo to a sweetheart.”" / "Trustpilot review" / "Your trainer shows you what to do when teeth land." / "GET YOUR CUSTOM TRAINING PLAN TODAY" / "80,000 owners and pets and counting".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1402 pixels tall): the hook lines are the biggest type, capitals about 64 pixels. The bubble quote about 40 pixels, the stars about 30 pixels, the resolution line about 28 pixels, the tag line about 24 pixels.

## CONCEPT
SOURCE 1249958710576916 is the layout to inherit: a pale seamless studio backdrop; a hand and forearm reaching in from the left edge toward the centre; a large white rounded speech bubble at the top right with a row of five green stars above a two-line quote in deep green and a tiny attribution under it; the object in the lower middle, softly shadowed. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Pale lilac-grey seamless #ECE9F0, soft top light.
- Top-left, 4% to 26%, left 52% of the width: the four hook lines in bold charcoal grotesk, flush left; 27% to 30%: "GET YOUR CUSTOM TRAINING PLAN TODAY" in small green capitals right under the hook, and under it "80,000 owners and pets and counting" smaller in grey.
- Top-right, 8% to 34%, right 44%: a white rounded speech bubble with its tail pointing down toward the puppy; inside, "★★★★★" in green #1F6B4F, the quote in deep green bold on two or three lines, and "Trustpilot review" small in grey.
- 38% to 86%: a real photograph on the same seamless: an adult hand and forearm in a cream knit sleeve reaching in from the left; a fluffy golden retriever puppy sitting on the floor, clamped onto the fingers with its tiny teeth, tugging, playful eyes, the sleeve cuff stretched.
- 90% to 93%: "Your trainer shows you what to do when teeth land." centred charcoal. Nothing else below.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>

Stories/Reels 9:16 · gpt-5.6-sol · high · attempt a1 · reference: 1249958710576916.jpg (the source ad) · exact text sent (Codex session log)

Use your built-in image_gen tool (not a CLI, not a script). Do not run repository scripts, do not edit files, do not copy files. GENERATE exactly one image at 941x1672 pixels (or the closest accepted size with the same aspect ratio; state which size it accepted).

MANDATORY TOOL ARGUMENT: call tools.image_gen__imagegen exactly once with referenced_image_paths set to this exact JSON array: ["/home/ubuntu/Documents/AdMaker/brands/dog-academy/output/statics-r2b-2026-09-28/sources/1249958710576916.jpg"]. Do not use num_last_images_to_include. The outer attachments alone are not sufficient.

## IDENTITY
Dog Academy sells in-home dog training, not a packaged product: there is no bottle, box, food pack, jar, app screen or product to show, so show none. Invent no logo, icon mark, badge, certificate or seal. Wherever the words "Dog Academy" appear they are plain type with no symbol beside them. People are anonymous everyday people: no name tags, no logos or printing on clothing, no uniforms. No trainer is presented as a named person.

## REALISM
A real, art-directed photograph: natural light with its real falloff, true skin and fur texture, correct dog anatomy (four legs, one tail, natural paws, ears and teeth), leashes that clip to a collar or harness, no CGI sheen, no plastic retouching. It should look better made than the source: one focal point, clean depth, a considered colour palette.

## RULES
LAYOUT: the top 16% of the frame is picture or background only, and the first word starts at 17% of the height or lower, below the phone's profile bar.
OUTPUT SIZE IS MANDATORY AND IS THE FIRST CHECK: a 9:16 image, exactly 941x1672 pixels (width 941, height 1672), a tall phone-screen shape. Do not return a square or 4:5 image.
SAFE ZONE: the top 14% and the bottom 22% of the frame hold picture only, no words at all (the phone's interface covers them). Every word sits between 15% and 76% of the height.
Create ONE complete, finished 9:16 Meta Stories/Reels ad for Dog Academy in one pass, a close copy of the attached source ad's design (the last attached image): its type placement, type weights, devices and colour logic, recomposed natively for a tall phone screen and made to look even more polished. Our scene, our words. Never copy the source's logo, brand name, product, faces, dogs or photographs.
Exact text only, spelled exactly as quoted, nothing else written anywhere: "Your puppy melts" / "your heart. Then" / "sinks teeth into" / "your hand." / "★★★★★" / "“My pup went from Cujo to a sweetheart.”" / "Trustpilot review" / "Your trainer shows you what to do when teeth land." / "GET YOUR CUSTOM TRAINING PLAN TODAY" / "80,000 owners and pets and counting".
No prices, no phone numbers, no web addresses, no watermarks, no dog tags with text, no books, signs, labels or printed props with words. No em dashes in any on-image text.
Exact sizes (the image is 1672 pixels tall): the hook lines are the biggest type, capitals about 60 pixels. Quote 36, stars 28, resolution 26, tag 22 pixels.

## CONCEPT
SOURCE 1249958710576916 is the layout to inherit: a pale seamless studio backdrop, a hand and forearm reaching in from the left, a large white rounded speech bubble with five green stars and a quote. The source is the last attached image. Change everything else — our scene, people, words and colours. Do not copy its words, faces, logo or pixels.
- Pale lilac-grey seamless #ECE9F0. 0% to 14%: plain seamless, no words.
- 15% to 29%: the four hook lines in bold charcoal grotesk, flush left; 30% to 33%: "GET YOUR CUSTOM TRAINING PLAN TODAY" in small green capitals, flush left, right under the hook, and "80,000 owners and pets and counting" smaller in grey under it.
- 33% to 46%: the white speech bubble at the right, tail pointing down, "★★★★★" in green #1F6B4F, the quote in deep green bold, "Trustpilot review" small in grey.
- 46% to 70%: the photograph: the adult hand and forearm in a cream knit sleeve reaching in from the left, the fluffy golden retriever puppy clamped onto the fingers, tugging the cuff.
- 72% to 76%: the resolution line centred charcoal. Nothing else below it.
- 78% to 100%: the seamless floor and the puppy's soft shadow, no words.


When finished, reply with exactly one line: ACCEPTED_SIZE=<WxH>