What is invisible moped AI?
Invisible moped AI, also called Crazy Frog AI or the invisible motorcycle trend, is a short AI video in which a person rides a scooter that does not exist. Nothing is filmed: an AI video model builds the whole clip from photos. There is no frog in it either; the name comes from the soundtrack.
The sound goes back a long way. In 1997 a Swedish teenager recorded himself imitating a moped engine; in 2003 the animator Erik Wernquist turned that recording into “The Annoying Thing”, and in 2005 its Crazy Frog version of “Axel F” became a hit around the world. In October 2026 people started re-making that frog’s ride with themselves: an early AI version was posted by @tonnatusha on 7 October, and the overpass chase that most remakes copy was posted by @wdmowski on 12 October.
The original chase has one rider and an anonymous cop. The two-person version swaps that cop for a friend, a partner or a coworker, which is what this page is for.
The two-person chase in one table
| Question | Answer on this site |
|---|---|
| What you upload | Photo 1, the rider, and Photo 2, the cop. Two separate selfies are fine; you do not need a photo of you together. |
| Who is who | Photo 1 crouches on the invisible moped in every rider shot. Photo 2 rides the white hover-bike in a grey police uniform and peaked cap over their own face. Swap flips them before or after a video. |
| Only one photo? | Switch to “An AI cop”: an invented adult officer who does not resemble the rider chases instead. |
| Length and frame | 15 seconds. 9:16 vertical (default) or 16:9 wide, same price. |
| Places | Overpass at dusk, neon city at night, supermarket aisle, or an empty photo of your own place. |
| Price | 99 credits (about $0.98) on Seedance 2.0 Mini at 480p, up to 1,757 credits on Seedance 2.5 at 1080p. Two photos cost the same as one. |
| Watermark | None, before or after a purchase. |
| If it fails | Every credit comes back automatically. |
Make a two-person invisible moped video from separate selfies
- Leave the generator on “Someone you know (2 photos)”. It opens that way on this page.
- Put the rider’s photo in Photo 1 and the chaser’s photo in Photo 2. Each photo shows one person (or one pet), face toward the camera, no sunglasses.
- Check the roles. If the wrong person is on the moped, press “Swap rider and cop”.
- Pick where the chase happens and the frame: 9:16 for TikTok, Reels and Shorts, 16:9 for a wide screen.
- Choose a model and resolution. Seedance 2.0 Mini at 480p is the cheapest way to check both faces first.
- Press “Make the invisible moped chase”. The tool first makes one still with both of you in the scene, then the 15-second video. Expect a few minutes; you can close the page.
- Download the MP4. To see it the other way round, use “Swap roles and make another”.
Who should ride and who should chase
The rider is on screen for most of the clip; the cop gets two seconds of close-up and is a small figure behind the rider the rest of the time. Cast with that in mind.
| Pair | Usually rides | Usually chases | Photo tip |
|---|---|---|---|
| Couple | Whoever the clip is for, or the one who will post it | The partner with the most serious frown | Two clear selfies in similar light, so the faces come out equally sharp |
| Birthday surprise | The birthday person | You, or the friend organising it | A waist-up or full-body photo of the birthday person; a suit or costume carries over to the ride |
| Coworkers | The colleague being celebrated or teased (with their OK) | The manager or the team lead | Avoid badges and lanyards with names on them |
| Siblings | The younger one, chased by the older one | The older one | Both photos of people 18 or over; see the consent section |
| Person and pet | Either one | A pet cop sits upright on the hover-bike, paws on the handlebars, no uniform | The whole animal, facing the camera, eyes sharp |
| Meme account, new pair daily | Today’s guest | A recurring host | Keep the host’s photo, place and frame the same every day so only the rider changes |
When each of you is on screen
Every video follows the same cuts as the clip the trend copies. This is where Photo 1 and Photo 2 appear, which helps when you check a result.
| Time | Shot | Who is on screen |
|---|---|---|
| 0–3.3 s | Low tracking shot beside the rider, crouched with fists forward | Photo 1 |
| 3.3–5.2 s | The rider glides toward the camera; the hover-bike appears far behind | Photo 1, Photo 2 tiny in the distance |
| 5.2–7.2 s | Close-up of the cop on the hover-bike, frowning into the lens | Photo 2 alone |
| 7.2–9.3 s | The rider glances back at the cop following close behind | Both |
| 9.3–10.9 s | Backflip in mid-air, landing back in the crouch with both hands up | Photo 1, Photo 2 behind |
| 10.9–15 s | Low wide shot of the getaway, then a sweeping shot from above | Both, small |
Keep two faces apart: what to check and fix
Two faces in one chase can go wrong in ways a solo clip can’t. Most of it comes from the photos, not the model.
| What you see | Likely cause | What to do |
|---|---|---|
| The wrong person is on the moped | The photos went into the wrong slots | “Swap roles and make another”, or Swap before you start |
| The cop in the close-up looks like the rider | The two photos look alike (same angle, light, glasses) or the cop’s face is small | Use a closer, front-facing photo of the cop with a plain background |
| One face looks like a blend of both | Two people in one of the photos, or a group photo cropped tight | One person per photo; crop each from a separate picture if needed |
| Glasses, beard or hairstyle changed | They are hidden or soft in the photo | A sharp photo where they are clearly visible; try Seedance 2.0 Pro or 2.5 if Mini drifts |
| The cop wears your friend’s own clothes | Expected: the uniform goes over their face, hair and build | Nothing to fix; a pet cop has no uniform at all |
| Blurry faces after posting | 480p is re-compressed by TikTok and Reels | Check the faces at 480p, then make the keeper at 720p |
This template vs Kling 4.0, Genjutsu and Seedance 2.5 for two people
General video models can make the same chase, but two people means you assign the roles yourself. This table is about what each route asks of you, from public descriptions checked on 21 October 2026; we did not run the same photos through each tool.
| Route | What you prepare | How the two roles are set | Where the chase motion comes from |
|---|---|---|---|
| This template | Two selfies | Fixed slots: Photo 1 rides, Photo 2 chases; Swap flips them | Built in: the trend’s cuts and timing, 15 s |
| Kling 4.0 (released 15 October 2026) | Reference images per person, a motion reference clip, a prompt | In your prompt; each person needs a clearly separate reference | A video you supply as a motion and camera reference |
| Higgsfield Genjutsu | A source clip and the character images to swap in | By swapping characters into the clip you bring | Your source clip’s own movement, camera and rhythm |
| Seedance 2.5 | Several reference images and an optional video, plus a prompt | By naming each reference image in the prompt | The prompt and any reference video; it reinterprets more freely |
Bringing someone else’s viral clip as a motion reference raises its own rights question; this template does not ask you for one.
A role-locked two-person prompt for other tools
If you use Kling, Seedance or Genjutsu yourself, give each person a separate reference image and say who does what in every shot. A starting point (replace Image 1 and Image 2 with your tool’s reference names):
Vertical 9:16, photorealistic, 15 seconds. The RIDER is the person in Image 1 and only that person; the COP is the person in Image 2 and only that person. They never trade faces. 0–5 s: the RIDER hovers in a seated crouch above an overpass road, knees bent, both fists forward on handlebars that do not exist; no moped, scooter or seat under them. 5–7 s: close-up of the COP alone on a white wheelless hover-bike with two round red front lights, grey police uniform and peaked cap, frowning into the lens. 7–9 s: the RIDER looks back at the COP behind. 9–11 s: the RIDER does one backflip and lands back in the crouch, hands up. 11–15 s: low wide shot of both racing away. No text, logos or watermark.
Keep the two reference images as different as you can: different backgrounds, both faces sharp and front-on.
What a two-person invisible moped video costs
Credits for one 15-second video, read from the generator. The second photo is free: solo and duo cost the same, and so does every place and frame.
| Model | 480p | 720p | 1080p |
|---|---|---|---|
| Seedance 2.0 Mini | 99 ($0.98) | 196 ($1.94) | — |
| Seedance 2.0 Fast | 195 ($1.93) | 404 ($4.00) | — |
| Seedance 2.0 Pro | 319 ($3.16) | 666 ($6.59) | 1,650 ($16.34) |
| Seedance 2.5 | 452 ($4.47) | 997 ($9.87) | 1,757 ($17.39) |
| MiniMax H3 | — | 448 ($4.44) | — |
Dollar figures use the Studio pack ($99 for 10,000 credits, $0.0099 a credit). Smaller packs cost a little more per credit.
Three pairs on Seedance 2.0 Mini at 480p: 297 credits. Ten pairs: 990 credits, inside the $19.90 Starter pack (1,600 credits).
Free credits, watermark and refunds
| Question | Answer |
|---|---|
| Is signing up free? | Yes, no card needed. A card or other payment is asked for only at checkout. |
| Do welcome credits make a video? | No. The few welcome credits some new accounts get are below the cheapest video (99 credits), and free credits only ever pay for Seedance 2.0 Mini. |
| Cheapest way to try | The one-time $4.99 trial pack: 99 credits, exactly one Seedance 2.0 Mini 480p video, solo or duo. |
| Watermark | None on any video. |
| Failed video | All credits back automatically, whether the still, the review or the video fails. |
| Subscription | None. Credit packs are one-time and last 365 days. |
Before you cast someone: consent, age and posting
- Ask first. Put someone’s face in the chase only if they agreed, even for a private group chat. Coworkers’ photos from a company website are not consent.
- Adults only. Use photos of people 18 or over. A younger brother or sister should not be cast, and posting AI videos of minors breaks most platforms’ rules.
- Label it. TikTok asks you to mark realistic AI-generated videos as AI-generated; turn the label on when you post.
- The sound. The video comes with the chase’s sound. For a brand post or an ad, the Crazy Frog name and the “Axel F” recording need their own licence; for a normal post, adding the sound from TikTok’s own library is the safer route.