le0xet

making content with AI | video, marketing, agents, & the prompts behind every clip

[le0xet @le0xet]

A GUY DROPS FOOD INTO LAVA FROM A HELICOPTER AND MAKES $12,800 A MONTH OFF THE COMMENTS

> 208,000 likes on 12,400 comments.

Open helicopter door over a cracked lava field, a man in a monkey mask hugs a cheese wheel the size of a tyre and lets it go.

It shrinks to a dot, sinks into a glowing pool, and a second later a column of molten cheese blasts straight back up and buries him in the doorway.

The return trip is doing more work than the drop, and that is the whole design:

-> Same door, same mask, same lava as the chicken-wing clip, so the viewer knows the bit and only waits to see what comes back.

-> The wheel hangs in the frame for a full second before it falls, a detail half the audience only catches on the second watch.

-> The cheese comes back as cheese, yellow, stringy and dripping, so it reads as a food joke and never as an explosion.

-> He stays in the door and takes it, tongue out on the mask, and never flinches like a stunt double.

-> The comments are "what cheese", "prompt?" and "do pizza next", and the last one is a series request.

A drop everyone can predict is a shortcut the brain cannot skip, and sending the object back up is the cheapest punchline in short-form.

It works on anything that should disappear and comes back instead, a watermelon, a bag of popcorn, a tub of paint, a birthday cake.

The word under it is PROMPT, the clip came out of @Picsart from one still of the helicopter door, Blotato sends the prompt behind an email gate, and Claude counts which food made people ask, $49 a month for both.

The cheese got the shares, and the door everyone knew from last week is what got them to wait for the splash.

One word under every drop, article below ↓

[le0xet @le0xet]

A GUY DROPS FOOD INTO LAVA FROM A HELICOPTER AND MAKES $12,800 A MONTH OFF THE COMMENTS

> 208,000 likes on 12,400 comments.

Open helicopter door over a cracked lava field, a man in a monkey mask hugs a cheese wheel the size of a tyre and lets it go.

It shrinks to a dot, sinks into a glowing pool, and a second later a column of molten cheese blasts straight back up and buries him in the doorway.

The return trip is doing more work than the drop, and that is the whole design:

-> Same door, same mask, same lava as the chicken-wing clip, so the viewer knows the bit and only waits to see what comes back.

-> The wheel hangs in the frame for a full second before it falls, a detail half the audience only catches on the second watch.

-> The cheese comes back as cheese, yellow, stringy and dripping, so it reads as a food joke and never as an explosion.

-> He stays in the door and takes it, tongue out on the mask, and never flinches like a stunt double.

-> The comments are "what cheese", "prompt?" and "do pizza next", and the last one is a series request.

A drop everyone can predict is a shortcut the brain cannot skip, and sending the object back up is the cheapest punchline in short-form.

It works on anything that should disappear and comes back instead, a watermelon, a bag of popcorn, a tub of paint, a birthday cake.

The word under it is PROMPT, the clip came out of @Picsart from one still of the helicopter door, Blotato sends the prompt behind an email gate, and Claude counts which food made people ask, $49 a month for both.

The cheese got the shares, and the door everyone knew from last week is what got them to wait for the splash.

One word under every drop, article below ↓

[Fokki @0x_fokki]

https://t.co/KqM5W6Kr01

[le0xet @le0xet]

WATERMELON ZOOMED OUT INTO FARMLAND

> 263,000 likes with 537 comments.

Macro on cut fruit, pulling back, and the pulp becomes a field seen from altitude. It keeps pulling back and lands exactly where it started.

Look at the engagement shape before you look at the visuals:

-> 263,000 likes against 537 comments, about one comment per five hundred likes, which is close to silence.

-> Like-heavy and comment-light is the signature of passive watching. Nobody had anything to say, they simply never left.

-> The loop is seamless, so the end is the beginning and a replay never announces itself.

-> Every one of those replays counts again as watch time, and watch time is what ranking actually weighs.

-> It runs eleven seconds, short enough that three loops feel like one viewing.

What gets manufactured here is repeat seconds, produced without the viewer ever deciding to watch again.

That matters commercially because platforms pay for exactly that metric in reach. An account that reliably generates watch time holds an asset it can point at anything later, and the unit cost is whatever the render cost.

Which is why loops survive every algorithm change. There is no loophole, the loop satisfies the metric honestly. The system asks for time spent and a loop supplies it, cleanly, forever.

The economics underneath shifted this year too. An infinite zoom through scales used to be a compositing job measured in days, and the scale transitions are now the part the model handles.

If you want to test a loop idea before committing to one, image-to-video off a single still is the shortest way in. @Picsart does it from a phone.

Nobody watched it three times. It just never ended ↓

[le0xet @le0xet]

WATERMELON ZOOMED OUT INTO FARMLAND

> 263,000 likes with 537 comments.

Macro on cut fruit, pulling back, and the pulp becomes a field seen from altitude. It keeps pulling back and lands exactly where it started.

Look at the engagement shape before you look at the visuals:

-> 263,000 likes against 537 comments, about one comment per five hundred likes, which is close to silence.

-> Like-heavy and comment-light is the signature of passive watching. Nobody had anything to say, they simply never left.

-> The loop is seamless, so the end is the beginning and a replay never announces itself.

-> Every one of those replays counts again as watch time, and watch time is what ranking actually weighs.

-> It runs eleven seconds, short enough that three loops feel like one viewing.

What gets manufactured here is repeat seconds, produced without the viewer ever deciding to watch again.

That matters commercially because platforms pay for exactly that metric in reach. An account that reliably generates watch time holds an asset it can point at anything later, and the unit cost is whatever the render cost.

Which is why loops survive every algorithm change. There is no loophole, the loop satisfies the metric honestly. The system asks for time spent and a loop supplies it, cleanly, forever.

The economics underneath shifted this year too. An infinite zoom through scales used to be a compositing job measured in days, and the scale transitions are now the part the model handles.

If you want to test a loop idea before committing to one, image-to-video off a single still is the shortest way in. @Picsart does it from a phone.

Nobody watched it three times. It just never ended ↓

[Valentin @martynov014]

https://t.co/I1Z0BwhrSi

[le0xet @le0xet]

THIS AI CLIP FAKES A MINIATURE DESERT SET, A CRANE CAMERA AND A FULL EFFECTS CREW IN 15 SECONDS

What sells it is the scale model, long before the monster. A tabletop village of tiny tents, sand dunes under stage lights, two operators in hi-vis vests at a monitor desk, and a green wall behind all of it.

For the first eight seconds it is a model helicopter on a wire doing a slow pass over the village.

Then the sand bulges, a giant worm breaks out of the dune, swallows the helicopter mid-air and sinks back while the crew keeps rolling.

That is the fake production logic that keeps working. One miniature, one practical-looking gag, one payoff the eye accepts because it happens at tabletop scale.

With @Picsart the same set goes through image, motion and edit and still reads as one shoot, with the crew and the lights holding between every cut.

One table of sand is good for a worm, a flood, a sandstorm and a crash, and that beats generating one epic desert shot every time.

The full pipeline with every prompt is in the article below ↓

[Valentin @martynov014]

https://t.co/I1Z0BwhrSi

[le0xet @le0xet]

@Solomonrojie 😅

[le0xet @le0xet]

I TOLD GROK BOT:

"Give me the cutest stupid AI video idea that could make me my first $1,000".

I expected a cat in a costume.

I got a claw machine full of live kittens.

A guy at an arcade drops the claw, it closes on one kitten, the kitten slides down the chute, and he pulls it out of the prize door and holds it like he just won it.

You understand the whole clip in one second and still watch to see if the claw lets go.

And that's basically the whole strategy:

> Put something alive where something plastic belongs.
> Keep the arcade, the lights and the guy completely real.
> Let the machine do what machines do.
> Give the payoff a face.
> End the second he lifts it out.

No story, no dialogue, one joke the eye gets before the brain does.

Grok handled the idea.

I ran it through @Picsart from my phone.

One dumb prompt, one 10-second clip, and now I want to see how far it goes toward that first $1,000.

The full formula with the prompt is in the article below ↓

[Lunar @LunarResearcher]

https://t.co/nN6fldtV4f