Crazy Frog AI chase · get it right the first time
Invisible Moped AI
Invisible moped AI is the October 2026 TikTok trend in which a real person, or a pet, crouches in mid-air on a moped nobody can see and races along an overpass while a police officer on a white hover-bike closes in; the rider looks back, backflips and gets away, usually to Crazy Frog's 2005 hit “Axel F”. The version almost everyone copies is the dusk overpass chase @wdmowski posted on 12 October 2026. This generator follows that clip shot for shot and puts your photo on the moped.
15-second MP4 · 16:9 or 9:16 · 480p, 720p or 1080p · 5 models · 98–1,740 credits per video (about $0.98–$17.40) · no watermark · failed videos refund their credits
Photo 1 · Rider (on the invisible moped)
One person or pet, face clear and facing the camera
Photo tips & requirements
300–6000 px wide and high, up to 10 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 · 16:9 · 15 s MP4 · the trend sound on the video
This video costs 447 credits. Charged when you start, refunded automatically if it fails.
Sign in to generateExample results
One photo. One wild chase.
Choose your cast
You, a friend or a pet. Pick the rider and the cop.
Set the scene
Overpass, neon city, supermarket — or your own place.
Download your MP4
A 15-second chase, in landscape or vertical.
What invisible moped AI makes, second by second
Every video here keeps the timing of the @wdmowski original, the first clip in the examples below. Knowing what each second shows tells you where a weak photo or a cheap model will give itself away.
| Time | On screen | What it needs from your photo |
|---|---|---|
| 0–3.3 s | Low tracking shot: you hover in a seated crouch, fists on handlebars that are not there, grinning into the curve. | A body to crouch with. A photo cut off at the chin leaves the model to invent your shoulders and clothes. |
| 3.3–5.2 s | You glide straight at the camera; the hover-bike appears far behind you. | The largest view of your face in the whole clip. Even light and no sunglasses count most here. |
| 5.2–7.2 s | Cut to the cop alone, frowning into the lens, two round red lights glowing. | Nothing from you. With two photos this is Photo 2's close-up; with one photo the cop is invented. |
| 7.2–9.3 s | Back on you: a glance over the shoulder at the cop, then eyes forward again. | A three-quarter view of your head. Glasses and hairlines are most likely to drift during this turn. |
| 9.3–10.9 s | A full backflip above the road, landing back in the crouch with both hands up and a shout. | The fastest movement in the clip. Hands, legs and faces soften here first, most of all at 480p. |
| 10.9–15 s | Wide shots: you shrink into the distance with the hover-bike behind, then a sweeping view from above. | Very little detail. At this distance the colour of your clothes is what keeps you recognisable. |
Invisible moped AI examples on the same chase
The first clip is the @wdmowski TikTok the template follows. The other three are results from crazyfrogai.org on the same beats and the same sound: a rainy neon street like this site's Neon city background, and two scenes you can recreate by uploading a background photo of your own. None of them was made on this site.
The original overpass chase
Source: @wdmowski on TikTok
Rainy neon city, denim jacket
Source: crazyfrogai.org
Pixel world, suit and tie (own background)
Source: crazyfrogai.org
Grandma in glasses, Halloween road (own background)
Source: crazyfrogai.org
Why your face survives the backflip: a still first, then the chase
This generator never animates your photo directly. It first draws one still of you on the invisible moped in the place you picked, with the cop behind you. The instruction for that still asks for your exact face proportions, eyes, hairline, skin tone and build, and rules out slimming, de-ageing or beautifying anyone.
Only then does a video model animate the still, with your photo as a second reference and a depth map of the @wdmowski chase for the motion and the cuts. Because the cuts come from that reference, the cop cannot drift into your shot and the flip cannot come early. What is left to the model is how much detail it holds at speed.
So a good photo does more than a bigger model: if the still looks like you, the video starts from you.
Photo checklist for invisible moped AI
JPG, PNG or WebP, 300–6000 px, up to 10 MB, one person or pet per photo. These are the photo choices that decide most results.
| Your photo | What tends to happen | Do this instead |
|---|---|---|
| Glasses | Thin frames can vanish or change shape when your head turns at 7–9 s. | Use a photo where the frames are clearly visible and not catching glare. Bold frames hold up better than rimless ones. |
| Curly or voluminous hair | The volume flattens in the flip when the photo crops the top of your head. | Leave space above your hair and pick a photo in even light, so the curls have texture to copy. |
| Cap, hood or hat | It stays on, and your hairline is guessed for the look-back. | Keep it if it is part of the joke; take it off if you want your own hair in the video. |
| Sunglasses | Your eyes are invented, so the front view reads as a stranger. | Use a photo without them. The upload hint asks for this too. |
| Selfie cut at the chin | Your clothes and body are invented and can change between shots. | A waist-up or full-body photo gives the crouch and the flip a real body to work with. |
| Group photo | The model may pick the wrong person or blend two faces into one. | Crop to one person per photo. For two people, use rider + cop mode with two photos. |
| Pet | Works as the rider or the cop; fur colours and markings are carried over from the photo. | Pick a photo where the whole animal is visible and facing the camera. |
Invisible moped AI problems and how to fix them
The seven complaints that come up most, why they happen, and what to change: on this generator, and when you build the chase yourself in Kling, Seedance or Higgsfield Genjutsu.
| Problem | Why it happens | On this generator | Doing it yourself |
|---|---|---|---|
| The face looks like someone else | A small or soft face gets rebuilt from a generic one, or the model animates a selfie with nothing anchoring who it is. | Upload a sharp, front-facing waist-up photo in even light; render the final version on Seedance 2.5 or 2.0 Pro at 720p. | Make a still of yourself in the scene first and animate that, with your original photo attached as a second identity reference. |
| Glasses or curls disappear | Fine detail is lost in the head turn and the flip, worst at 480p. | Choose 720p or 1080p and a photo where the frames and hair are clearly visible. | Name the glasses and the hairstyle in the prompt and repeat them for every shot. |
| A real motorcycle appears underneath | Video models have seen countless riders on bikes and fill the empty space. | The still and the motion reference both leave the space empty. If a bike shows up anyway, contact support from your account with the video. | Write that there is no motorcycle, scooter, seat, wheels or frame under the rider, and describe fists gripping empty air. |
| Hands open, legs change position | With text alone the model improvises the pose between shots. | The depth reference holds the crouch and the fists for the whole chase; a full-body photo keeps the legs consistent. | Use a motion reference (Kling Motion Control, Genjutsu Motion Transfer) instead of describing the pose in words. |
| The flip comes too early, or the cop is in the wrong shot | Text prompts describe the beats loosely, so models reorder them. | Cannot happen here: the cuts at 5.2, 7.2, 10.9 and 12.8 s come from the reference clip. | Write the prompt with time codes, keep the reference clip's own cuts and replace only the rider. |
| Two friends swap faces | Two faces without a fixed role get mixed when the camera cuts. | Rider + cop mode gives each photo one role: Photo 1 rides, Photo 2 is the cop, and Swap flips them. | Give each person their own reference image and name their role in every shot. |
| Blurry on TikTok | A 480p file stretched by the app, often after a second export from an editor. | Make the posting version at 720p and upload the MP4 exactly as downloaded. | The same: generate at 720p or higher and post the original file. |
Which model to choose for invisible moped AI
All five run the same chase from the same reference. They differ in resolution, how much detail survives the flip, and credits.
| Model | Resolutions | Credits per video | Pick it when |
|---|---|---|---|
| Seedance 2.0 Mini | 480p, 720p | 98 (480p) · 194 (720p) | You want to check whether a photo works before paying for a sharper version. |
| Seedance 2.0 Fast | 480p, 720p | 193 (480p) · 400 (720p) | You want 720p for a casual post at a lower price. |
| Seedance 2.0 Pro | 480p, 720p, 1080p | 315 (480p) · 659 (720p) · 1,633 (1080p) | You want 1080p for a little less than Seedance 2.5. |
| Seedance 2.5 | 480p, 720p, 1080p | 447 (480p) · 987 (720p) · 1,740 (1080p) | The default. It reviews your photos before it starts; use it for the video you will actually post. |
| MiniMax H3 | 720p | 444 (720p) | You want a different look from the Seedance family; 720p only. |
480p, 720p or 1080p for TikTok?
TikTok, Reels and Shorts show vertical video at up to 1080 × 1920, but most people watch on a phone held at arm's length. 720p is where the face in the look-back and the hands in the flip stay readable after the app compresses your upload.
Use 480p to test a photo or for a quick joke in a group chat. Pay for 1080p only if you will crop, zoom or show it on a big screen. On Seedance 2.5, going from 480p to 720p adds 540 credits; from 720p to 1080p adds 753.
Making invisible moped AI yourself in Kling 4.0, Seedance 2.5 or Genjutsu
If you would rather control every detail, these are the routes people use (checked 19 October 2026; these tools change fast), followed by a prompt written to keep the bike invisible.
| Route | What you give it | What keeps the chase in order | What holds your face | Effort |
|---|---|---|---|---|
| This generator | One or two photos and a background | A depth reference of the @wdmowski clip; the cuts are fixed | A still of you is made first, then animated | Upload, choose, wait a few minutes |
| Kling 4.0 or Kling Motion Control | Your photo plus a reference video for motion and camera | The reference video you supply | Your photo and prompt only | Find a reference clip, write a prompt, retry |
| Seedance 2.5 reference mode | Images and a reference video, called up in the prompt | Your prompt and the reference; it reinterprets freely | Whatever stills and photos you attach | Prompt writing; good for changing places or characters |
| Higgsfield Genjutsu Motion Transfer | A reference video and your character | The reference's own motion, camera and timing | Your character replaces the person in the reference | Most setup; closest to the original movement |
A prompt that keeps the bike invisible
9:16 vertical, 15 seconds, photoreal. The person in the reference image rides an invisible moped: hovering in a seated crouch just above a concrete overpass at dusk, knees bent, feet off the ground, both fists gripping empty air where handlebars would be. Nothing is under them: no motorcycle, scooter, seat, wheels or frame. 0–5 s: low tracking shot beside the rider, then the rider glides toward the camera, grinning; a police officer on a white hover-bike appears far behind. 5–7 s: close-up of the officer only, stern, two round red front lights. 7–9 s: the rider glances back over one shoulder, then faces forward. 9–11 s: the rider backflips above the road and lands in the same crouch, hands raised, shouting with joy. 11–15 s: wide shots as the rider speeds away with the hover-bike following. Keep the same face, hair, glasses and clothes in every shot. No text, captions, logos or extra people.
With a reference clip, add: “Follow the reference video's motion, positions, cuts and camera exactly; replace only the rider.” For two people, describe each person once together with their role (“the woman in image 1 is the rider; the man in image 2 is the officer”) and keep those roles in every shot.
Price, free credits, watermark and refunds
The short answers, read from the same price list checkout uses.
| Question | Answer |
|---|---|
| Is there a free invisible moped AI video here? | No. There is no free generation; the cheapest way to make one video is the $4.99 trial. |
| What does the trial include? | 450 credits for one successful video, for example Seedance 2.5 at 480p (447 credits). First purchase only, one per account. |
| Credit packs | Starter $9.90 / 900 · Plus $24.90 / 2,400 · Pro $49.90 / 4,990 credits. One-time payments, no subscription; credits last a year. |
| Watermark | None. Videos download as clean MP4s. |
| If a video fails | Every credit for a failed video goes back to your balance automatically. Making a finished video again costs credits again. |
| Do I need a card? | You need a payment method to buy credits; checkout shows the methods available in your country. |
Crazy Frog AI price per video
Every video is 15 seconds. Credits depend only on the model and resolution: 16:9 and 9:16 cost the same, one photo or two cost the same.
| Model | Resolution | Credits | ≈ $ on the $49.90 pack | ≈ $ on the $9.90 pack |
|---|---|---|---|---|
| Seedance 2.0 Mini | 480p | 98 | $0.98 | $1.08 |
| Seedance 2.0 Mini | 720p | 194 | $1.94 | $2.13 |
| Seedance 2.0 Fast | 480p | 193 | $1.93 | $2.12 |
| Seedance 2.0 Fast | 720p | 400 | $4.00 | $4.40 |
| Seedance 2.0 Pro | 480p | 315 | $3.15 | $3.47 |
| Seedance 2.0 Pro | 720p | 659 | $6.59 | $7.25 |
| Seedance 2.0 Pro | 1080p | 1,633 | $16.33 | $17.96 |
| Seedance 2.5 | 480p | 447 | $4.47 | $4.92 |
| Seedance 2.5 | 720p | 987 | $9.87 | $10.86 |
| Seedance 2.5 | 1080p | 1,740 | $17.40 | $19.14 |
| MiniMax H3 | 720p | 444 | $4.44 | $4.88 |
Packs: $4.99 trial (450 credits, first purchase only), $9.90 (900), $24.90 (2,400), $49.90 (4,990). Credits last 12 months. Failed videos are refunded in credits automatically. No subscription.
Invisible moped AI questions
Why doesn't my invisible moped AI video look like me?
Usually because of the photo: a small, soft or side-lit face gives the model too little to copy, so it fills in a generic one. Use a sharp, front-facing photo from the waist up in even light. Here your face is fixed in a still before the chase is animated, so if the still looks right, the video starts from you.
I wear glasses and have curly hair. Will more photos or a different model help?
This tool takes one photo per person, so pick your clearest one: frames visible, hair not cropped. Then make it at 720p or higher on Seedance 2.5 or 2.0 Pro. Fine detail is lost first at 480p, in the head turn and the backflip.
Why does a real motorcycle sometimes appear under the rider?
Video models have seen countless people sitting on bikes and fill the gap. On this generator the still and the motion reference both leave the space empty. In your own prompts, say outright that there is no motorcycle, scooter, seat, wheels or frame, and describe the fists gripping empty air.
Can the backflip come too early, or the cop show up in the wrong shot?
Not on this generator: the cuts at 5.2, 7.2, 10.9 and 12.8 seconds come from the reference clip, not from a text prompt. In Kling or Seedance, write the prompt with time codes and keep the reference clip's cuts, replacing only the rider.
My friend and I keep swapping faces. How do I stop it?
Give each face exactly one role. In rider + cop mode Photo 1 always rides and Photo 2 is always the cop; Swap flips the roles without mixing the faces. When you prompt a model yourself, attach one reference image per person and name the role in every shot.
Is 720p or 1080p worth it for TikTok?
720p is worth it for anything you post: the face and the flip stay readable after TikTok compresses the upload. On Seedance 2.5 that is 987 credits instead of 447 at 480p. 1080p only pays off if you crop, zoom or show it on a large screen. Upload the MP4 as downloaded rather than exporting it again from an editor.
What happens to my credits if a generation fails?
All of them go back to your balance automatically. If a video finishes but something is clearly wrong, such as a missing rider, contact support from your account and include the video.
If keeping my face matters most, should I use a one-click invisible moped AI generator or a general model?
Start with a one-click generator that builds a still of you first and has the chase timed in advance; there is less for the model to get wrong. General models such as Kling, Seedance or Genjutsu give you more control over place, story and ending, in exchange for more prompt work and more retries.
Crazy Frog AI generator
The same chase on the homepage, with the song, the trend timeline and rider/cop roles.
Is Crazy Frog AI free?
Every way to make it by cost, from free CapCut templates to the trial.
Crazy Frog AI on your phone
Make it in Safari or Chrome and save the MP4 to Photos or Gallery.
Pricing
One-time credit packs and the credits each model costs.
Put your photo on the invisible moped
One clear photo is enough. Pick the model and resolution from the tables above and let the chase run.
Back to the generator