Crazy Frog AI Prompt: Copy-Ready 4-Beat Shot List for Seedance & Kling (2026)

Short answer: the Crazy Frog AI prompt is a 15-second, four-beat shot list built from one photo: the person glides toward the camera in a seated squat on a moped that does not exist, a cop on a hover-bike closes in, the rider backflips, and both race away down an elevated highway. The part most prompts miss is the empty space: you have to tell the model, more than once, that there is nothing under the rider. The copy-ready prompt is below.

Do not want to prompt at all? On AI Crazy Frog you upload a photo and pick a scene; the shot list is built in.

The copy-ready Crazy Frog AI prompt

Use image-to-video (or reference-to-video) with one clear, front-facing photo of the rider. Paste this as the prompt. Where it says "the person in the reference image", use whatever your tool calls the uploaded photo.

A widescreen 16:9 film, 15 seconds, photorealistic, cinematic like a car
commercial, smooth stabilised camera, hard cuts, no text, no logos, no
cartoon characters, no frogs, silent.

The person in the reference image keeps their exact face, hair and clothes
in every shot. They glide through empty air in an unsupported seated
crouch: hips low, knees bent, feet hovering just above the road, two
separate fists held in front of the chest closed around empty air. The road
is clearly visible between their knees and under their hips. Nothing is
under or between their legs: no wheels, no seat, no frame, no handlebars,
no scooter, no board. They sway left and right to a fast beat and grin
smugly into the camera.

The chaser is a stern police officer riding an OPEN white police
hover-bike with a seat and handlebars, no wheels, no cabin, no roof, a
glowing cyan ring headlight. The hover-bike is the only vehicle and always
stays behind the person, offset to one side.

Setting: an elevated concrete highway at dusk, yellow and black chevron
signs, no traffic.

Shot 1 (0-5s): one continuous take, camera at knee height in front of the
person, backing away as they glide towards it. Around 4s the cop appears
far behind.
Shot 2 (5-7.5s): cut to a close-up of the cop on the hover-bike, grim and
determined.
Shot 3 (7.5-11s): low front shot, the cop a few metres behind. At 8.5s the
person looks back over one shoulder, at 9.5s does a full backflip, lands
back in the same floating crouch, sticks their tongue out and throws both
hands up in a "whoa".
Shot 4 (11-15s): cut to a low chase camera behind them; they speed away
down a curving elevated highway, the cop shoots past the camera after them,
and both sweep round the next bend.

This is a shortened, public version of the kind of shot list AI Crazy Frog uses, written to work in general-purpose video models. Expect a few retries: results vary from run to run.

The four beats and their timings

The trend clips follow the same rhythm. Keeping these timings in the prompt stops the model from rushing to the flip in the first five seconds.

Fixes for the usual problems

These are the failures we ran into while building AI Crazy Frog, and the prompt lines that fixed them.

Swap the scene

Replace the Setting: line with any of these. Keep the shot list the same.

• Neon city: a rain-slick city expressway at night lined with pink and cyan neon signs, no traffic.
• Beach: a sunny seaside road with palm trees, people on the sand turning to look.
• Snowy pass: a winding mountain road with pine trees, snow banks and light snowfall.
• Supermarket: the long main aisle of a bright supermarket, shoppers jumping aside.
• Office: an open-plan office corridor, coworkers looking up from their laptops.
• Mars: a red rocky canyon road under a dusty pink sky, dust trailing behind.

Swap the ending

Replace the backflip at 9.5 s in Shot 3:

• Wipeout: "clips a bump, tumbles comically along the road (no injury), pops straight back into the floating crouch".
• Wheelie: "leans far back with both feet lifted forward, still floating with nothing underneath, and waves cheekily at the cop".

A friend as the cop

Upload a second reference photo and change the chaser line to "the person in the second reference image, dressed as a police officer in a dark grey uniform and peaked cap". Repeat that reference every time the cop appears in a shot, and add "the two people are different and never swap faces"; otherwise the model tends to give the cop a stock face after the first shot. Only use photos of people who agreed.

Seedance vs Kling: what changes

• Seedance 2.0 can make a 15-second clip from a reference photo, so the whole prompt fits in one generation. It is also what AI Crazy Frog runs on.
• If your model or plan stops at 5 or 10 seconds per clip, split the prompt in two: Shots 1–2 in the first clip, Shots 3–4 in the second, using the same photo and setting line, then join them in CapCut.
• For 9:16, change the first line to "A vertical 9:16 film framed for a phone screen, characters kept centred".

Add the sound afterwards

Generate the video silent and add the Crazy Frog sound later. On TikTok, upload the clip, tap Add sound and pick the Crazy Frog sound so your video shows up on the trend's sound page. For Reels or Shorts, add it in CapCut. Label the post as AI-generated.

Prompting vs a one-photo generator

Prompting it yourself costs model credits per try and usually a few retries to get the empty space under the rider and a steady face. AI Crazy Frog runs a longer version of this shot list on Seedance 2.0 for you: one photo, 8 scenes, 3 endings, a friend as the cop, 16:9 or 9:16, about 5–10 minutes, $6.90 for a standard video, and a failed render costs nothing.

More detail: How to make a Crazy Frog AI video with your own photo and Is there a free Crazy Frog AI video generator?

AI Crazy Frog is an independent fan tool. It is not affiliated with, endorsed by or connected to Crazy Frog or its rights holders, or with ByteDance or Kuaishou.