Invisible moped AI · routes compared
Invisible Moped AI: Genjutsu, Kling 4.0 or a one-photo template?
Invisible moped AI is the Crazy Frog AI trend: a photo becomes someone crouched on a moped that isn't there, racing along an overpass while a cop on a hover-bike gives chase, then a look back, a backflip and a getaway to Crazy Frog's “Axel F”. Below: which tool keeps your face and the original beats, what each one costs, and a template that makes it from one photo.
Make the chase from a photo- 1 photo, or 2 with a friend as the cop
- 15 s · 9:16 or 16:9
- Failed runs refunded in credits
- From 27 credits
Photo 1 · Rider (on the invisible moped)
One person or pet, face clear and facing the camera
300–6000 px wide and high, up to 30 MB. One person or pet per photo, facing the camera in even light, no sunglasses. Waist-up or full-body photos give the crouch and the flip a real body to work with. Adults only — use your own photo, or one from someone who said yes.
480p · 9:16 · 15 s MP4 · the trend sound on the video
This video costs 121 credits. Charged when you start, refunded automatically if it fails.
Chase examples
Made on other sites; choose a clip to inspect the action. These are not your results.
Which route to pick, in one line each
Start from the outcome you care about. Each row is explained in the comparison table further down.
| Route | Pick it when |
|---|---|
| One-photo template (this page, crazyfrogai.io, crazyfrogai.org and similar) | You want the chase people recognise from one selfie today, with no clip to find and no prompt to write. |
| Higgsfield Genjutsu (Motion Transfer, Object Swap) | You have a chase clip you're allowed to use and want its exact moves and cuts with a new rider. |
| Kling 4.0 / Kling Motion Control | You want to try the newest reference-video model on your own footage or a clip you have rights to. |
| Seedance 2.5 (via Dreamina or CapCut Pro) | You'd rather reinterpret the chase — another place, other effects — than copy it shot for shot. |
| PixVerse app | You only want a phone app and can live without a ready-made Crazy Frog preset. |
| CapCut “CRAZY FROG” templates | You already filmed yourself crouching and only need the Axel F audio and cuts; CapCut won't draw the chase. |
What invisible moped AI is — and why there's no frog
Invisible moped AI is what many people type when they mean the Crazy Frog AI trend that spread across TikTok and Reels in October 2026. Nobody in it is a frog: a real person, or a pet, is lifted into a seated crouch with fists on handlebars that don't exist and sent down a highway with a futuristic police bike behind. Invisible motorcycle AI, invisible bike AI and Crazy Frog AI all name the same clip.
The frog lives in the soundtrack. The Annoying Thing, Erik Wernquist's 2003 animation built on a Swedish teenager's 1997 impression of a moped engine, became Crazy Frog, whose cover of “Axel F” topped charts in 2005. The AI clips borrow that track and the joke of an engine noise with no engine.
Early AI versions appeared in the first week of October 2026. The overpass chase most remakes copy — cop close-up, look back, backflip, getaway — was posted by @wdmowski on 12 October 2026, and the template on this page follows its timing.
Six ways to make it, compared
Each row describes how the tool handles this particular chase, checked on its own pages. Credits don't compare across sites, so prices stay in each tool's own units, with dollars only where a site sells in dollars or publishes a rate.
| Route | What you upload | Where the moves come from | What holds your face | Length and frame | One clip | Free use and watermark | Best for |
|---|---|---|---|---|---|---|---|
| One-photo template (this page, crazyfrogai.io, crazyfrogai.org and similar) | One photo, or two with a friend as the cop; optionally a background photo. | A built-in reference of the chase. You bring no clip and no prompt. | Your photo as an image reference. This tool also turns it into a scene still first. | This tool: 15 s, 9:16 or 16:9. crazyfrogai.io: 15 s, 9:16. crazyfrogai.org: 11 s, 16:9. Frog Race AI: 12 s. Viberot: 8 s. | This tool: from 27 credits (≈ $1.00). crazyfrogai.io $6.99 at 720p; crazyfrogai.org $9.90; Frog Race AI $4.99; Viberot 110 credits at 720p (≈ $1.82). | crazyfrogai.io: one free 480p preview, watermarked. Viberot: 20 free credits a month, fewer than a clip needs. crazyfrogai.org: no free tier. This tool: see the free table below. | The chase people recognise, from a selfie, with no setup. |
| Higgsfield Genjutsu (Motion Transfer, Object Swap) | A 4–30 s source video plus up to 30 reference images; text is optional. | Your clip. Motion Transfer keeps its motion, camera and timing; Object Swap replaces one subject. No Crazy Frog template. | Your reference images. | Up to 30 s, up to 1080p. | Higgsfield's own table: 104 credits ($5.20) for 15 s at 720p, 40 credits at 480p. Starter is $19 a month for 270 credits. | Free accounts get limited use and every free generation is watermarked; paid plans aren't. Plus and Ultra include two or three free Genjutsu runs. | Recasting a clip you're allowed to use, move for move. |
| Kling 4.0 / Kling Motion Control | Kling 4.0: text, images, keyframes or up to five reference videos (30 s in total). Motion Control: one character image and one motion video. | Your reference video (4.0 also takes white-mannequin and depth-map clips), or a prompt. | Your images as references. | Kling 4.0: 3–30 s, 720p to 4K, 16:9, 9:16, 1:1 or 21:9. Motion Control: as long as your reference. | Kling 4.0's per-second rate isn't published. Motion Control on VIDEO 3.0 Standard is 9 credits a second: 135 credits for 15 s, about $2.05 at $1 = 66 credits. Standard plan: $6.99 the first month, then $8.80 for 660 credits. | The free plan gets promotional credits only, and free output carries the Kling watermark. | Trying the newest reference-video model on footage you own. |
| Seedance 2.5 (via Dreamina or CapCut Pro) | Text, images, video and audio; up to 50 references per clip. | A prompt or a reference video; it reinterprets more freely than a straight motion copy. | Your images as references. | Up to 30 s per generation; Dreamina's beta long-video mode extends further. | CapCut lists Seedance 2.5 from $0.12 a second on Pro (first month), about $1.80 for 15 s. Dreamina showed local-currency prices only. | Rolled out to subscriber accounts; no free-use or watermark rule is published for 2.5. | Re-imagining the chase with another place or other effects. |
| PixVerse app | A photo or text, or one of the app's templates. | A prompt or an app template; no Crazy Frog preset found. Copying motion from a video is documented for the developer API, not the app. | Your photo as the starting image. | 540p free, 720p on Standard, up to 4K from Pro; clip length isn't listed. | The per-second cost isn't listed. Standard is $10 a month for 1,200 credits plus daily credits; packs start at $5 for 500. | 60 starting credits plus 30 a day, up to 540p, watermarked. Standard and above are watermark-free. | Phone-only makers willing to write their own prompt. |
| CapCut “CRAZY FROG” templates | Your own clips or photos. | Your footage, timed to the music. No AI chase is generated. | Your own footage, unchanged. | The most-used “CRAZY FROG” templates run 13–14 s, in 9:16 or 3:4. | The templates are free; CapCut Pro's price varies by region. | Free templates on the free plan; template watermark rules aren't published. | Adding Axel F and cuts to a crouch you filmed yourself. |
Kling 4.0 launched on 15 October 2026, according to Kuaishou. Higgsfield's 15-second price is from its own Genjutsu post. One-click sites differ from each other too: crazyfrogai.io adds a wipeout ending and a two-rider option, and Frog Race AI swaps the cop for a chasing car.
Checked on 2026-10-19 on each tool's own pages. Prices and free tiers change; check before you pay.
Which route keeps your face
What decides it is whether your photo reaches the model as an image reference or only as words. Every route in the table takes a reference image; none can promise your face survives a fast backflip.
Your photo as a reference
Template sites, Genjutsu, Kling and Seedance all take your photo as an image input. Faces still drift when the head turns away, at speed, or when the photo is small, filtered or behind sunglasses.
A description only
Describing yourself in a prompt (“a man with curly hair and glasses”) gives you a stranger who fits the description. Keep prompts for invented characters.
What this tool sends
Your photo first becomes one still of you in the chosen place; the model then gets that still, your original photo and the reference chase. On Seedance 2.5 the photos and the clip pass the provider's asset review before rendering. In duo mode the cop's photo goes in as its own image, so the two roles don't share a face.
Test any route with the same photo
We have not run one photo through every tool, and no price table can tell you how your face comes out. A fair test checks the same moments in each result:
- 0–1 s: is it you — face shape, glasses, hair and clothes?
- About 8 s, the look back: does the face hold when the head turns?
- About 10 s, the flip and landing: count arms and legs, and check the face isn't smeared.
- Throughout: nothing appears under the rider — no scooter, seat or wheels.
- 5–7 s: the cop's close-up sits mid-clip, not at the start or the end.
- With two people: after every cut, each face stays with its own role.
Which route hits the original beats
The chase people recognise has seven shots. A template keeps them because its motion is fixed; reference-video tools keep them only when you give them a clip; a prompt has to spell out each one.
| Time | Shot | What to watch |
|---|---|---|
| 0–3.3 s | Crouched on the invisible moped along the overpass | Fists forward, knees bent, feet off the road, no vehicle. |
| 3.3–5.2 s | Racing toward the camera, hover-bike far behind | The cop should still be small in the background. |
| 5.2–7.2 s | Cop close-up on the hover-bike | A close-up in the wrong place is the most common prompt-only mistake. |
| 7.2–9.3 s | The rider glances back over a shoulder | Where faces most often change. |
| 9.3–10.9 s | Backflip and hands-up landing | Extra limbs, a flip that cuts off, or a bike appearing at the landing. |
| 10.9–12.8 s | Low wide getaway | Rider and cop both in frame, the rider pulling away. |
| 12.8–15 s | Sweeping aerial shot | Some generators drop it; the clip still reads as the trend. |
Template (this tool)
A depth map of the overpass chase drives every model and its audio becomes the soundtrack, so the shots arrive in order and on the beat. The price is a fixed ending.
Genjutsu, Kling 4.0, Kling Motion Control
They take the timing from a clip you upload. That means finding the original or filming your own take, and having the right to use it.
Prompt only (any model)
Write the seven shots with timestamps and still expect the cop or the flip to land in the wrong place on some runs. Budget for re-runs.
Inside this tool: five models, one chase
Every model gets the same reference chase and returns a 15-second MP4 with the Axel F audio. Credits are per whole video and the same for solo or duo and for either frame.
| Model and quality | Credits | ≈ USD at the Pro monthly rate | When to use it |
|---|---|---|---|
| Seedance 2.5 · 480p | 121 | $4.50 | The default. Reviews your photos and the clip before rendering; 480p to 1080p. |
| Seedance 2.5 · 720p | 266 | $9.89 | The default. Reviews your photos and the clip before rendering; 480p to 1080p. |
| Seedance 2.5 · 1080p | 468 | $17.40 | The default. Reviews your photos and the clip before rendering; 480p to 1080p. |
| Seedance 2.0 Mini · 480p | 27 | $1.00 | The cheapest way to see how a photo comes out before spending more; 480p or 720p. |
| Seedance 2.0 Mini · 720p | 52 | $1.93 | The cheapest way to see how a photo comes out before spending more; 480p or 720p. |
| Seedance 2.0 Fast · 480p | 52 | $1.93 | A quicker Seedance 2.0 at 480p or 720p. |
| Seedance 2.0 Fast · 720p | 108 | $4.02 | A quicker Seedance 2.0 at 480p or 720p. |
| Seedance 2.0 · 480p | 85 | $3.16 | Up to 1080p without the 2.5 review step. |
| Seedance 2.0 · 720p | 178 | $6.62 | Up to 1080p without the 2.5 review step. |
| Seedance 2.0 · 1080p | 440 | $16.36 | Up to 1080p without the 2.5 review step. |
| MiniMax H3 · 720p | 120 | $4.46 | A different model family, 720p only — worth a run if Seedance keeps changing a face. |
Dollar figures convert credits at the Pro monthly rate ($29 for 780 credits), the lowest per-credit price we sell monthly. What you pay depends on the plan or pack you buy.
Solo or duo
One photo makes you the rider and invents the cop. Switch to duo to cast a friend — or a pet — as the cop; each photo keeps its own role.
Four places
The overpass, a neon city, a supermarket aisle, or a background photo you upload, such as your shop or street.
What stays fixed
The ending is always the backflip and getaway, the length is 15 seconds, and there is no slot for two riders side by side. If you need those, see when a general model is worth it below.
Free credits, watermark and refunds by route
A free account rarely finishes a whole chase. This is what each route gives before you pay, and what happens when a run fails.
| Route | Free credits, and enough for one MP4? | Watermark | Card at sign-up | Failed run |
|---|---|---|---|---|
| This page's tool | 3 gift credits; the cheapest chase needs 27. Gift credits can only fund Seedance Mini. Gift validity: 7 days. | Before a first paid order, generated videos carry the site watermark. A paid account can download the saved clean original when available. Downloading an existing result does not use more generation credits. | No card needed to sign up. | A failed task returns its credits automatically. A finished video you dislike counts as a completed run. |
| Higgsfield Genjutsu | Free use is “limited”; the amount isn't published, so don't count on finishing a 15 s clip. | Every free generation is watermarked; paid plans are not. | Not published. | Credits come back automatically, usually within minutes. |
| Kling | No daily credits on the free plan, only promotional ones. | Free output is watermarked; removal is a paid benefit. | Not published. | Failed generations are refunded; a Motion Control result that comes out shorter is not. |
| PixVerse | 60 credits at sign-up plus 30 a day, up to 540p. | Watermarked on the free plan. | Not published. | No published rule for failed runs; credit packs are non-refundable. |
| Dedicated one-click sites | crazyfrogai.io: one free 480p preview. Viberot: 20 credits a month, below the 51 its cheapest clip needs. crazyfrogai.org: none. | Paid videos on crazyfrogai.io and crazyfrogai.org have no watermark; crazyfrogai.io's free preview does. | Viberot's free credits need no card; the others don't say. | crazyfrogai.io, crazyfrogai.org, Frog Race AI and Viberot all return credits or charges for failed videos. |
Checked on 2026-10-19 on each tool's own pages. Prices and free tiers change; check before you pay.
What 1, 10 or 30 clips cost by route, re-runs included
Plan for re-runs: here every second keeper needs one more try, so 10 clips means 15 runs. This tool is shown in credits with the single purchase that covers them; sites that sell per video are shown in dollars at their listed price.
| Route | 1 clips (2 runs) | 10 clips (15 runs) | 30 clips (45 runs) |
|---|---|---|---|
| This tool · Seedance 2.5 · 480p | 242 creditsStarter pack · $17.00 once | 1815 creditsEnterprise plan · $99.00/month | 5445 creditsMore than any single purchase |
| crazyfrogai.io15 s · 9:16 · 720p; packs of 1, 5, 35 or 90 videos | $13.98 | $87.00 | $157.00 |
| crazyfrogai.org11 s · 16:9; 1, 3 or 10 videos | $19.80 | $113.70 | $320.70 |
| Frog Race AI (crazyfrog-ai.com)12 s · 720p · a car chases you; 1 or 3 videos | $9.98 | $49.95 | $149.85 |
| Viberot8 s · 720p (110 credits); packs of 600 to 11,000 credits | $9.90 | $29.70 | $69.80 |
Each site's figure is the cheapest mix of its own packs that covers the runs. Only crazyfrogai.io's clips are as long as this tool's 15 s. Genjutsu, Kling and PixVerse sell monthly credits whose cost per clip depends on plan and model, so they stay in the routes table.
Checked on 2026-10-19 on each tool's own pages. Prices and free tiers change; check before you pay.
When a general model is worth the setup
A template is fastest when you want the trend as people know it. Reach for Genjutsu, Kling or Seedance when you need something a template can't do.
A different ending
Wipe out, get caught, ride into the sea: this template always flips and escapes. Put the new ending in a prompt, or film it as your reference.
Two riders side by side
This tool has one rider and one cop. Two friends racing side by side need a general model, or a site with a two-rider option, with one reference image per person and each role named.
Your own crouching video
If you filmed yourself pretending to steer and want your own moves kept, that's a video-to-video job. On this site, Replace Person in Video takes your clip and portraits.
Replace a person in your own videoSomeone else's viral clip
Feeding a stranger's TikTok into a reference-video tool reuses their footage and their face, and no tool's terms give you rights to it. Use your own take, or a template that ships its own reference.
Four chases from one-click generators
Each clip comes from another site and follows the same chase this tool uses — crouch, cop close-up, look back, flip, getaway — with the Axel F audio. They show what the template route looks like; they are not results from this page.
The closest to the original: the same overpass and camera moves, with the cop played by a second person — what duo mode is for.
Source: MixVioA pet rider in a neon city. The paws stay on the invisible handlebars and the cop's close-up lands mid-clip.
Source: aicrazyfrog.videoThe chase moved into a bright hall from an uploaded photo. The cop arrives about two seconds early and the aerial ending is missing — check those beats on your own run.
Source: Remake Your MemeIt starts on an overpass and becomes a pumpkin-lit road after the cop's close-up: a custom scene can change mid-clip.
Source: Remake Your MemeQuestions about choosing a route
Is Higgsfield Genjutsu better than Kling or Seedance for the exact same moves?
For copying a clip move for move, Genjutsu's Motion Transfer is built for the job and Kling 4.0 now takes a reference video too; Seedance 2.5 tends to reinterpret more. No one has published a same-input comparison, so run one photo through two of them and compare the frames at 0, 8 and 10 seconds. If you don't have the clip, a template already carries the moves.
Do I need the original TikTok as a motion reference?
Only for reference-video tools such as Genjutsu or Kling Motion Control, and then you also need permission to use it. A one-photo template brings its own reference: this one uses a depth map of the overpass chase, so you upload a photo and nothing else.
Can I feed someone else's viral clip into Genjutsu or Kling?
The tools will accept the upload, but their terms leave the rights with you, and the clip carries another person's footage and face. For anything you post publicly or use for a business, start from footage you shot or a template whose reference comes with the tool.
Can CapCut alone make the floating ride?
No. CapCut's Crazy Frog templates time your photos or clips to the Axel F audio; they don't generate you crouching in mid-air or the cop chasing you. Generate the chase first, then edit it in CapCut if you like.
One-click template or my own prompts — which keeps my face closer?
Both depend on your photo going in as an image reference. A template removes the choreography problem, so you judge the face alone; prompts add variables. Use a sharp, front-facing photo without sunglasses or filters, and compare the look-back and flip frames, not only the first second.
Why did my friend's face swap with mine when the camera cut?
The model lost track of which reference belongs to which role. Give each person a separate photo, name the roles in order (rider, then cop) and avoid two people with similar hair or clothes. In duo mode here, Photo 1 is always the rider and Photo 2 the cop.
I run a meme page — how do I keep the chase the same with a new person every day?
Keep the reference fixed instead of writing a fresh prompt each day: a template, or one reference clip you own in Genjutsu or Kling. Keep the model, quality and frame the same, change only the photo, and note the settings with each post.
Credits for your next chase
Buy credits once or monthly; the same credits work on every template on this site.
Starter
For trying the trend
240 credits / month
≈ 10 ten-second Seedance 2.0 Mini videos
- Watermark-free videos, including ones you already made
- Every model, length and resolution
- Failed videos refunded
- Email support
Pro
PopularFor regular creators
780 credits / month
≈ 32 ten-second Seedance 2.0 Mini videos
- Watermark-free videos, including ones you already made
- Every model, length and resolution
- Failed videos refunded
- Priority support
Enterprise
For agencies and pages
2,650 credits / month
≈ 110 ten-second Seedance 2.0 Mini videos
- Everything in Pro
- Dedicated support