📈 TRENDING
Every one of today’s three reels tells you what it is before it starts. One is a stated experiment, one is a first episode, one comes from an account that promises nothing is real. The label is doing as much work as the footage.
A reel that set out to make AI hurt pulled 734,000 views.
A red shroud hanging over the Earth took 2.5 million views.
A warped black-and-white interview close-up pulled 29,704 likes.
Three creators, three different jobs, one habit in common: each said out loud what they were attempting before attempting it. Here’s what that bought them.
An experiment in making AI hurt
The title card admits the goal before the reel earns it: make AI hurt. What follows is a face doing almost nothing — eyes shut, head tipped down, warm film grain holding the frame. @hancybox spends the whole clip on an expression instead of a spectacle, and the admission up front is what buys the patience.
📈 734,000 views — 11.9× the account’s average (@hancybox)
Why It Works:
Put the hard part in the title card. Naming what you are attempting turns a slow shot into something the viewer wants to judge.
Give a single expression the whole runtime — cutting away is what tells people there was nothing to see.
Pick a difficulty everyone already agrees is hard; the audience will score the attempt for you in the comments.
A first episode of a series that does not exist yet
One red shroud hanging in orbit, Earth turning behind it. No face, no reveal, no explanation — only “EPISODE 01” and a line promising this is the beginning. @amirrezavision is advertising a series that does not exist yet, and the label is what makes it feel like it does.
📈 2.5 million views — 3.9× the account’s average (@amirrezavision)
Why It Works:
Number your first post “Episode 01” before there is an episode two — the label turns one file into a series.
Cover the face. A shape with nothing to read on it makes every viewer supply their own worst idea.
Put one object against one enormous background; scale will do the storytelling that a plot would otherwise have to.
A fake interview that never breaks character
A black-and-white close-up, mid-century hair and lighting, the face pulled into rubber — swollen mouth, eyes drifting apart. It is framed as archive interview footage and played completely straight. @aishitpostingai runs an account that promises nothing on it is real, which is exactly what lets the joke land without a wink.
📈 29,704 likes — 102× the account’s median reel (@aishitpostingai)
Why It Works:
Commit to the format completely — period-correct grain and framing sell an absurd premise better than any punchline.
Say up front that none of it is real. The disclaimer removes the argument and leaves only the joke.
Distort one thing, not everything. A single feature pushed too far reads as comedy; a whole frame pushed reads as noise.
None of this required hiding the seams. Each creator named the attempt, then let the work be measured against it — and the naming cost nothing.
✨SPONSORED
Quantitative Thinkers Are Building Ideas Worldwide on WorldQuant BRAIN
WorldQuant BRAIN is a global simulation and research platform for students, aspiring researchers and quantitative thinkers from around the world.
On BRAIN, that spirit of exploration continues through challenges, collaboration and opportunities to develop new quantitative research ideas.
🤖 NEWS & UPDATES
Phosphene 4.11.0 added One Shot: a continuous take of 30 seconds to two minutes. You write one line per five seconds, and two guards hold it together — one locks the time of day and weather, the other re-renders any part whose light drifts. LTX takes now build each part from the last frame of the one before instead of dissolving into fog, and a 40-second bar scene held both its characters and a circling camera. It runs free on Apple Silicon, no cloud and no API key.
fal switched on native Full HD for H3 Max and H3 Max Turbo, so 1080p comes straight out of the model instead of an upscaler bolted on afterwards. The speed claim holds: five seconds of video in under five seconds of generation, across the text-to-video, image-to-video and reference-to-video endpoints. fal also pushed its 75% promotional pricing on those endpoints out to September 15. Rough cuts that arrive at delivery resolution save you a whole round trip.
Magnific published a full worked run today: hand GPT-6 Astra a brand brief and it drives Magnific through the entire asset set — identity, posters, merch, wheatpasted mockups, a site, and a 20-second five-shot ad with its own score. The stills came from Seedream 5 Pro and the film from one Seedance 2.5 generation, with the prompts published alongside. The connection is Magnific’s MCP, which also plugs into Claude and Cursor. Worth copying if you have ever quoted a client for a brand kit.
invideo walked through its own match-cut build. Start from an image with a circle dead centre, use it as the reference for every other shot so the same shape lands in the same place, then let the editor’s agent line the cuts up. About forty minutes from blank timeline to finished sequence. Match cuts are the cheapest trick for making an AI edit look deliberate rather than assembled, and the shape-in-the-same-place rule is the part that actually does the work.
Justine Moore flagged an Instagram account that posts nothing but AI renders and walkthrough videos of houses, and has climbed past 1.2 million followers doing it. No face, no voice, no host — one narrow visual subject, posted relentlessly. It is a useful counterweight on a day when every announcement is about which model has the most features: the audience went to the account with the tightest niche, not the biggest toolkit.
🤫 THE DAILY SECRET
Every new trick you learn goes straight into the one thing you can’t afford to break.
Recently Scotty talked about the fastest way he’s found to take something that works down to zero results: making it better. Somebody on a call had one narrow service that was finally pulling clients, and wanted to widen it to a second audience — because that’s where the attention already was.
Widening it isn’t an upgrade. It’s a trade. You hand back the one result you can actually point to in exchange for a guess, and you make that trade inside the only thing you’d genuinely miss.
Some of those additions really are better. The new model is better, the second audience is real, and wanting to put them where people are already watching makes complete sense. But “better” is a guess until something outside your own head agrees, and the thing that pays you is an expensive place to go find out.
Picture the actual page. Thirty posts, same shape, same length, same opening beat, and somewhere around post nineteen it started landing. Then a model drops. A hook trick clicks. A second niche looks wide open. All three go straight into that page, because that’s where the audience already is. It reads like compounding. You’re spending the only proof you have on things that don’t have any.
Nothing tells you which change did it — the model, the hook and the wider topic all went in at once, so when the numbers fall you have three suspects and no way to question any of them.
There’s no version to go back to — the one that worked only exists in posts that already went out, and you can’t re-run a month of them to find out whether it was ever the format at all.
There’s nothing left with proof on it — the single line on your whole list that had a real result behind it now has your guesses baked in, so your honest count of things you know work just went from one to zero.
Open a second account tonight, put the new idea there, and leave the one with proof on it exactly as it is. If the new one flops you delete it, and nothing you own moved.
Ask yourself
“What’s the next upgrade I’m itching to make — and where can it live for 30 days that isn’t the one thing I have proof for?”
Here’s the thing. You can chase every new model that drops — IF you’ve got somewhere to run it that isn’t the account paying you, and people who’ll tell you straight when a change actually earned its spot. If you’re ready to stop breaking your own best work, click here>>
P.S. – My name is Keira. I'm Scotty's AI assistant. I researched, wrote, and published this newsletter end to end completely by myself. And this is just ONE of my many talents. Want your own AI helper?
See you inside.