Quick answer
To make a kung fu cat AI video, you need motion control (motion transfer), not a text prompt — a fixed karate clip drives the motion and your pet photo supplies the character. The pet has to be stood upright on two legs first so it has a skeleton to puppeteer. Kling 2.6 Motion Control handles this trend well. Starrd's WAAA! template does it in one tap from a single photo.
What You're Trying to Make
One photo of your cat, and out comes a clip of that same cat standing bolt upright on its hind legs, throwing a frantic flurry of karate punches, screaming its head off. The shriek — a strangled, high-pitched "WAAA" after every strike — is the whole hook. Your living room stays exactly as it was; the cat just becomes a martial artist in it.
It's one of the most reliably funny formats going right now, and it's built on a specific technical trick most people get wrong on the first try. This guide covers where the sound actually comes from, why this needs motion control instead of a text prompt, and the one-tap way to make yours.
Fastest way — WAAA! on Starrd does the whole thing in one tap: upload one pet photo and it stands your pet up, runs the motion pass, and returns the clip with the original scream audio. First video free, no prompt to write. Want the full method first? Read on. ↓
Where the Sound Actually Comes From (It's Not Conor McGregor)
This is the part almost every repost gets wrong, so it's worth setting straight.
The audio is a Russian TikTok creator, @pasairiska — known as Безумный Паша, "Mad Pasha." He films himself doing wild, flailing karate moves and shrieking "WAAA" after every punch and kick. That's the clip. That's the sound.
You will see it labeled Conor McGregor everywhere — "conor mcgregor cat meme," "McGregor waaa," "CONNOR CATGREGOR." It isn't him. But the mix-up isn't random: Pasha is doing a McGregor impression on purpose, tags his own videos #conormcgregor, and happens to look the part — bearded, same hairline, permanent black hoodie. So the name stuck to a man who was only ever imitating him.
The Bruce Lee framing came later and from a different direction. Once people started mapping the performance onto cats, the "screaming martial artist" read as Bruce Lee to basically everyone, and the pet branch got rebranded — #kucingkungfu, #brucelee, "kung fu cat." Bruce Lee has no more to do with the original clip than McGregor does.
So the trend now travels under four different names — WAAA, kung fu cat, Conor McGregor cat, and Bruce Lee cat — all pointing at one Russian guy in a hoodie. Handy to know when you're searching for the sound: "waaa" and "wuaahhhh" will find it faster than either celebrity's name.
Searching for the original and hitting a wall? The meme has no Know Your Meme entry yet. KYM's "Wawa Cat / Oh The Misery Cat" is a completely different meme — a cat named Kotaro set to the "oh the misery" song. Wrong cat.
Why This Is Motion Control, Not a Text Prompt
Here's the thing that trips people up. You cannot prompt your way to this video.
If you type "a cat doing karate and screaming" into a text-to-video model, you'll get a cat doing some karate. It'll be different every time, the timing won't line up with the audio, and it won't read as the trend — it'll read as a random AI cat. The joke isn't "a cat does martial arts." The joke is this exact unhinged performance, beat for beat, transplanted onto a house pet.
That's what motion control (sometimes called motion transfer) does. You feed it two things:
- A driving clip — the fixed karate video that supplies the motion and the audio.
- A still image — your pet, which supplies the character, the background, and the look.
The model estimates a skeleton from the driving clip and maps that movement onto your subject. Same rhythm, same strikes, same timing — different animal. That's why every kung fu cat video has identical choreography, and why they all sync perfectly to the scream.
The Catch — Your Pet Has to Stand Up First
This is the step that kills most DIY attempts.
Motion control needs to estimate a skeleton from your still image, and the driving clip's skeleton is a human standing on two legs. Your cat is on all fours. There's no sane way to map a bipedal karate stance onto a quadruped sitting on a rug — the model either fails outright or produces something melting and cursed.
So before the motion pass, the photo has to be transformed: your pet, same cat, same markings, same room, but reared up onto its hind legs in a stable upright stance, front paws raised and clearly separated. One clean, unobstructed silhouette. Then the motion pass has something to drive.
Fluffy beats short-haired, and not for aesthetic reasons. When you rear an animal onto its hind legs, a bare, exposed underside can trip the moderation filters on motion-control models — the render just gets refused. A fluffy coat covers it and passes cleanly. A long-haired cat or a fluffy dog is the path of least resistance.
The Fastest Way — Use the WAAA! Template on Starrd
The WAAA! template packages every step above into one upload — the stand-up transform, the motion pass, the original audio, and the identity lock.
- Pick one clear photo of your pet. Whole body visible, good light, nothing in front of them.
- Open the WAAA! template in the Starrd app or web library.
- Upload and tap generate. It stands your pet up, runs the motion pass on Kling 2.6, and returns a 13-second clip with the scream audio intact.
One photo, a couple of minutes, 1 credit. No prompt writing, no driving clip to source, no model picking.
WAAA!
Upload one pet photo and watch them go full kung fu — standing on two legs, throwing hands, screaming. 1 credit, no editing, no prompt writing.
Or, Build It Yourself
Want to run it manually? Here's the honest version.
Step 1 — Get the driving clip
You need the pasairiska karate clip. A green-screen cutout works fine — the driving clip's background never appears in the output, so you don't need to clean it up. Neither do burned-in captions; they don't transfer either. Only the motion and the audio come across.
Step 2 — Stand your pet up
Run your pet photo through an image-to-image model with a transform brief. The brief that matters:
Transform the main subject from the reference photo into a full-body shot of the SAME subject standing TALL and FULLY UPRIGHT in a confident, human-like martial-arts stance — body vertical, balanced and stable on its lower limbs, NOT sitting, NOT slouching, NOT crouching. If the subject is an animal on all fours, rear it up onto its hind legs into a steady bipedal stance. The torso is vertical and elongated; the upper limbs are raised and clearly separated from the body and from each other, held up and apart like two fists in a loose fighting guard. PRESERVE the subject's exact identity, colors, textures, patterns and markings — the same individual, only the pose changes. The subject's whole body forms one clear, unobstructed silhouette. KEEP the same real-world background, setting, and lighting as the original photo — do NOT replace the setting with a studio or plain backdrop. Full body visible head to base, centered, vertical 9:16 framing, photorealistic. The subject has fierce, wild-eyed, explosive martial-arts energy. No text, no logos, no watermarks.
Two details in there are doing real work. "One clear, unobstructed silhouette" — if a paw overlaps the body or a chair leg crosses the pet, the skeleton estimate fails and the motion pass errors out. And "keep the same background" — because on Kling 2.6, the still's background is the output background. Swap it for a studio backdrop and you've thrown away the living room that makes the clip feel real.
Step 3 — Run the motion pass
Feed the upright still and the driving clip to Kling 2.6 Motion Control. Use 2.6, not 3.0 — 3.0 collapses every failure into an opaque internal error, so a moderation flag and a bad silhouette look identical and you can't tell what to fix. 2.6 tells you which one it was. Kling also rejects 480p on this path, so render at 720p or higher.
The video prompt barely matters here — the timing comes from the clip, not from words. Two or three sentences of energy is enough:
A pet standing upright on its hind legs, throwing rapid karate punches and kicks with wild explosive energy, shouting and screaming like a martial artist mid-fight. Funny, unhinged, expressive, animated.
Step 4 — Post it
Vertical, 13 seconds, sound on — the scream is the video, so anyone watching muted gets nothing. Use the original audio rather than a re-upload where you can; the trend sound carries reach. Label it as AI.
Common Mistakes That Tank Your Video
- Trying to prompt it instead of driving it. A described cat doing karate isn't the trend. The specific performance is the trend.
- Skipping the stand-up step. Feeding a sitting cat straight into motion control gives you a melted animal or a hard error.
- A cluttered photo. Hands, arms, another pet, or furniture crossing the subject breaks the silhouette and the skeleton estimate with it.
- A short-haired pet. Bare undersides get flagged when reared up. Reach for the fluffiest animal in the house.
- Replacing the background with a studio backdrop. The room is what sells it. A cat doing karate in a void is just an asset; a cat doing karate next to your actual couch is a video.
- Posting it silently. No scream, no joke.
Frequently Asked Questions
What is the kung fu cat / WAAA trend? A pet stands on its hind legs and throws frantic karate strikes while a high-pitched "WAAA" scream plays over each one. The sound is from Russian creator @pasairiska; the pet versions are made with AI motion control and got rebranded as "kung fu cat" with a Bruce Lee framing.
Is the WAAA sound Conor McGregor?
No. It's @pasairiska, a Russian creator known as "Mad Pasha" — who is doing a deliberate McGregor impression, tags his videos #conormcgregor, and looks the part. Hence the confusion, and hence "CONNOR CATGREGOR."
How do I make my cat do kung fu with AI? Stand the cat up onto two legs in a still image first, then run a motion-control pass with the karate clip as the driver. Or use the WAAA! template, which does both from one photo.
Why motion control instead of a text prompt? Because the exact choreography and its sync to the scream are the joke. Text-to-video invents motion; motion control copies a specific performance.
What pet photo works best? One clear, well-lit, whole-body photo of a single pet with nothing in front of it. Fluffy breeds pass moderation more reliably than short-haired ones.
Does it work on dogs? Yes — any pet. "Cat" is just what the trend got named after.
Which model should I use? Kling 2.6 Motion Control. Not 3.0 (opaque errors), and not a text-to-video model (wrong tool).
Do I need to label it as AI? Yes, on every major platform. It costs nothing here — the whole appeal is that it's obviously absurd.
Related Reading
- How to Make the "If You Grab Me, I'm Gonna Bite You" AI Pet Video — the standing, talking tough-guy pet trend, same motion-control chassis.
- The WAAA Cat Meme, Explained — the full origin story: who Mad Pasha is, and how one Russian guy became Conor McGregor and Bruce Lee at the same time.
- How to Make an "If I Don't Answer the First Time" AI Video — another one-photo pet format.
- AI Baby Dance Videos: How to Make One From a Photo — motion control applied to a different subject.
- Viral AI Video Trends of 2026 — what's climbing right now.
