Land the Setup Fast
The premise has a two-second budget. Anything longer and the clip turns into a sketch nobody finishes.
An AI artificial intelligence funny video generator for six-second bits: one setup, one punchline, and a face that holds when the laugh lands.
WRITE THE BIT









Prompt Guide
Model
Reference Image
Three decisions that decide whether a clip gets a second watch or a scroll.
Two jokes in six seconds is zero jokes. Choose the reveal and cut everything that delays it.
Medium shot, eyes to camera, nothing across the eyes. If the punchline lives in the expression, the expression has to survive a phone screen.
End on the reaction, not after it. Half a second of a held face lands harder than two seconds of laughing.
Pets, office fails, family reactions.
A large dog climbs onto a dining chair the moment a man stands up, sits neatly, and looks at the camera with ears back, static medium shot, warm dining room light, deadpan framing, 16:9.
Try This StyleA woman in a home office talks with her hands on a video call, unaware her microphone is muted, a colleague waves on screen, she keeps going, static laptop angle, morning window light, 16:9.
Try This StyleA birthday cake slides slowly off a table onto a tiled floor as a family watches, one child lunges and misses, an adult covers their mouth, handheld wide shot, warm party light, 16:9.
Try This StyleA pigeon in a tiny knitted scarf stands on a train platform as commuters walk past, it turns its head to follow a passing train, static low shot, cool fluorescent light, straight-faced realism, 16:9.
Try This StyleA cat on a sofa slowly turning its head away as someone off frame calls it, deadpan timing, static medium shot.
Try This PromptAn office worker staring into a jammed printer, tray hanging open, fluorescent light, locked medium shot, no dialogue.
Try This PromptAn older woman leaning very close to a laptop webcam, nose magnified, kitchen behind, static shot, warm light.
Try This PromptA dog sitting perfectly still, ears up, waiting, then one small head tilt, living room floor, eye level.
Try This PromptA spoon balanced on a nose for one second before sliding off, plain backdrop, harsh key light, tight framing.
Try This PromptA man frozen mid sentence on a laptop screen, cursor visible, bedroom desk, screen recorded look, flat light.
Try This PromptTiming, faces and physics are where AI comedy breaks. Fix these first.
The premise has a two-second budget. Anything longer and the clip turns into a sketch nobody finishes.
A half-second freeze on the face does more work than any sound effect.
Build the joke around invented faces. A public figure or a friend's face brings a takedown, not laughs.
Flat front light keeps the expression legible. Deep shadow turns a laugh into a squint.
Give the character something to hold. Empty hands drift and warp during fast reactions.
The last frame should be the punchline, not a camera move. Do not drift out.
Notes from people testing one joke at a time.
The printer clip beat everything I filmed myself last month. The pause before he closes his eyes is the whole joke, and I only got it on the third go by asking for a held expression.
I made a birthday video for my sister and dropped the cake slide in as a joke. She thought it was real for about two seconds, which is exactly the amount of time I wanted.
I write corporate training and we cannot film in a real office. A deadpan meeting scene at this cost is worth more to me than a polished sketch, and the flat lighting actually helps.
It renders what you describe, so the premise is still your job. Write the setup in one line and let the model build the room.
Fast expression changes leave too few frames. Slow the reaction and hold the last beat.
No. Invented faces avoid likeness problems and usually animate better than a real photo.
Yes. Grok Imagine Video 1.5 gives one free render.
Yes, if the licence allows commercial use and you label it as AI. Do not build the joke on a real person.