The most-watched new ad character of September 2026 was selling a memecoin.
His name is Jean Philanthrope. Jean Phil to the internet. Blond curled bob, handlebar moustache, brown houndstooth suit. He shadowboxes in European parks and never says a word.
He is not real. He first posted on September 17. Two days later a clip of him shadowboxing in a park did 5.6 million views on TikTok in three days. Within two weeks he had around 230,000 Instagram followers. Bored Panda counts 33 million views across his TikToks.
His bio pointed to a Solana memecoin, $JEANPHIL. It peaked on September 21 and lost more than 99% of its value the same day.
Strip out the coin and it is a better mascot brief than most brands write.
Who is Jean Phil?
An AI-generated persona. @jean_philanthrope on Instagram and TikTok, @JeanPhilMadame on X. Nobody has been named as the creator. No person, no agency, no brand.
What is known is the account's history. Jeremy Carrasco, who tracks AI accounts, put it flatly: "This account has seven videos. They changed their usernames six times in 15 months."
Then came the copycats. Archibald Brown, a British "rival" who challenges Jean Phil to a fight, with his own token and about 100,000 followers. Derek Mercer, with another. Dexerto summed up the format in one line: "an instantly recognizable character, repeatable videos, an imagined rivalry, and a token fans can buy."
One look. One bit. Zero to a quarter of a million followers in under two weeks.
Why weird AI characters work in ads
The weird gets you the first two seconds. The repetition gets you the memory.
System1 ran a study with TikTok across 92,000 people, 887 ads and eight markets. A recognisable character on screen in the first two seconds lifted brand awareness 57%. Creator-led ads got 39% more attention.
Over years it compounds. System1's work on the IPA Databank found campaigns with a recurring character or device are 37% more likely to grow market share and 30% more likely to grow profit.
And at the Super Bowl, where brands pay the most for a face, character-led ads averaged 3.7 stars in System1's scoring. Celebrity ads averaged 2.8.
A celebrity costs millions and leaves. A croissant in sneakers costs a prompt and works every Monday.
Brands already tried the borrowed version
The 2025 Italian brainrot wave was the first round. Ryanair posted a "Tralalero Tralala" TikTok. KFC pretended to put brainrot toys in its kids meal. Scrub Daddy and the Empire State Building's TikTok joined in. Liven, a mental health app, ran a fully AI-generated ad.
All of them borrowed characters they did not own. A borrowed character dies with the meme.
The play now is the reverse. Build the weird character yourself, own it, and run it for a year.
The AI ads are selling characters too
The AI video companies are running the pitch in their own paid social.
Runway on Meta: "Go from a single image to an AI influencer you can endlessly generate across any scene, situation or story." HeyGen runs creator-style ads that open on "This AI tool just took my website and turned it into a whole video with my AI character." Arcads opens with a synthetic presenter asking "I'm AI-generated, but I look real, right?"
Substack went the other way. "Everyone's got a case of brainrot. Dr. Randa is out here prescribing the cure: real writers, on Substack."
Their live ads are in the library: Runway, HeyGen, Arcads and Substack.
Searches for "ai slop ads" are up more than 7x on last year. People can smell it. Slop is a new random face in every ad. A mascot is the same face every week. Same tools, opposite result.
Build your own instead of borrowing his
We made three for this post. Crumb, a bakery croissant who has worked the 5am shift for 30 years. Gloria, a pigeon in a trench coat who judges your card statement for a budgeting app. Monty, a monstera on legs for a plant delivery brand.
Every character here came out of OpenAI's image model from the exact prompt printed next to it. Paste the same prompt into ChatGPT and you get the same kind of result. Nothing retouched.
Step 1: design the character
One prompt. Name, size, body, clothes, one personality trait you can see, and how it is shot. "Shoot it like a real photo" is what makes it weird instead of cartoon.
Design an original mascot for a neighborhood bakery's TikTok ads. His name is Crumb.
Crumb is a butter croissant the size of a house cat, standing upright on two skinny legs in white tube socks and red high-top sneakers. Two short stubby arms. Small round wire glasses. Deadpan, slightly tired expression, like he has worked the 5am shift for 30 years. Realistic flaky golden pastry texture with a few crumbs falling off.
Shoot it like a real photo, not a cartoon or 3D render: 50mm lens, soft daylight, shallow depth of field. Plain warm-grey studio background. Full body, centered, square image.
Step 2: lock it with a reference sheet
This is the step everyone skips and the reason their character has a different face in every ad. Same chat, one follow-up:
Turn this character into a character reference sheet I can reuse for video. Three views side by side on one plain warm-grey background, same lighting and same scale in every view:
1. Front view, head and upper body, looking straight at the camera.
2. Three-quarter view, full body, turned 45 degrees to the left.
3. Full body side profile, mid-stride, walking to the right.
Keep every detail identical to the image: shape, colours, textures, clothes, accessories and proportions. Photoreal, like the original. No text, no labels, no logos, no grid lines.
That sheet is the character now. Every video model that takes reference images, Seedance included, gets this sheet every time.
What makes one last
- A silhouette you can read at thumbnail size. A croissant. A pigeon. A plant.
- One bit. Jean Phil shadowboxes. That is the whole act.
- On screen in the first two seconds, not at the end card.
- The product inside the bit. Crumb is the bakery. Gloria is the budget.
- The same sheet every week, so week ten looks like week one.
Want a weekly pipeline of these finished?
Designing the character takes an afternoon. Shipping it every week is the job.
Director is a full creative team for your Meta ads. Every Monday you get finished, on-brand video ads, reviewed by a human. You pick the winners. Sprites ships the campaigns. Apply for Director.
For what is working right now, the hooks are in ad hooks that stop the scroll, the formats in video ad examples, and every brand's live ads in the Sprites ads library.
Frequently asked questions
Who is Jean Phil?
Jean Phil is Jean Philanthrope, an AI-generated character who went viral on TikTok and Instagram in September 2026. He has a blond curled bob, a handlebar moustache and a brown houndstooth suit, and he silently shadowboxes in European parks. His bio promoted the $JEANPHIL memecoin, which lost more than 99% of its value on September 21.
Is Jean Phil a real person?
No. Jean Phil is an AI-generated persona. No person, agency or brand has been named as his creator, and the account had changed its username six times in 15 months before he went viral.
Do AI commercials work?
A recurring character does. System1 and TikTok found a recognisable character in the first two seconds lifted brand awareness 57%, and System1 found campaigns built on a recurring character are 37% more likely to grow market share. A new random AI face in every ad gets none of that.
What are AI slop ads?
AI slop ads are cheap AI-generated ads with no consistent character, story or brand: a different face and a different style every time. Viewers recognise them and scroll. A mascot is the opposite, the same character every week until people know it.
Are brand mascots coming back?
Yes. Duolingo, Hotels.com and Mucinex all lean on characters, and in System1 Super Bowl scoring character-led ads averaged 3.7 stars against 2.8 for celebrity ads. AI makes a mascot cheap enough for a small brand to run one every week.
How do I make a consistent AI character?
Design the character in ChatGPT with one detailed prompt, then ask it for a reference sheet: front, three-quarter and full body on one plain background. Feed that sheet to your video model as the reference every time, so week ten looks like week one.
Jean Phil's coin is worth nothing now. The format isn't.