How to make a Crazy Frog AI video
Make a crazy frog AI video from two photos on MiniMax H3: 15 seconds, vertical or horizontal, sound included. Step-by-step, no subscription.
Last updated: 2026-10-15
Two photos and a few minutes is all it takes. The trend is one fixed chase — a rider squatting on a moped that isn't there, a chaser closing in behind, a flip and a getaway — and on MiniMax H3 you don't need the original clip or a reference video. This guide covers every step: which photos work, how to set the shape, where the sound comes from, what it costs, and how to fix a render that comes out wrong.
The short version: the chase in 6 steps
- Pick the rider. Photo 1 is the person squatting on the invisible moped — they flip in mid-air and speed off. Use yourself or a friend who said yes.
- Pick the chaser. Photo 2 rides behind and gets the close-up. Both photos are required.
- Add the two photos. One person each, face fully visible, facing the camera.
- Pick the road. Image 3 is an empty outdoor road — a preset, or a still you upload with nobody in it.
- Set the shape. Vertical for Shorts and Reels, horizontal for YouTube and X. Length is 15 seconds.
- Render, download, post. Usually a couple of minutes, then download the MP4 — sound is already in it. You do not write a prompt.
Step by step: making it on this site
Step 1 — Open the maker
The maker sits at the top of the home page, next to sample clips so you can watch what you'll get before you spend anything. It runs in a phone or laptop browser; there's no app.
Step 2 — Add the rider (Photo 1)
Tap the first slot. This is the face the model copies — the rider squats, pulls the look, flips and escapes.
- One person, facing the camera, face fully visible, even light, no sunglasses or masks
- Waist-up or full-body if you want the outfit to read at speed
- JPEG, PNG or WebP, up to 10 MB
- Faces the model may refuse: celebrities, cartoon characters, anyone under 18, and anyone who hasn't agreed
Step 3 — Add the chaser (Photo 2)
The second photo rides behind you and takes the close-up. Both faces stay consistent for the whole clip. The maker needs both slots filled — it will not start with Photo 2 empty. Same photo rules as Photo 1.
Step 4 — Pick the road (Image 3)
The chase prompt is already in the maker; you never type it. What you pick here is the empty road MiniMax H3 restyles the chase onto.
- Use one of the outdoor-road presets, or upload your own wide still
- Nobody in the frame, no cars, no shop-front text
- Halloween graveyard is first in the grid; the overpass is the closest match to the trend clip people remake
The full Image 1 / Image 2 / Image 3 wording is on the prompt page.
Step 5 — Set the shape
This is the part most generators don't let you touch.
- Length: 15 seconds (50 credits at 768P).
- Shape: vertical 9:16 for TikTok, Reels and Shorts; horizontal 16:9 for YouTube and X. The clip is rendered to that shape instead of being letterboxed afterwards.
Step 6 — Set the sound
H3 generates the audio in the same pass as the video: engine, road, ambience, and the beat your cut lands on. There is no separate silent switch in the maker. If you'd rather use the trend sound itself, lay it on inside TikTok (see The sound below).
Step 7 — Make the video and wait
Tick the box confirming both people agreed, then render. Usually a couple of minutes; 15-second clips take longer. You can close the tab — it keeps rendering and shows up under Generation history. There is no email when it finishes.
Step 8 — Download and post
Download the MP4 from the player. Post to TikTok, Reels or Shorts, mark it as AI-generated where the app asks, and add #crazyfrog or #crazyfrogai.
The three settings most generators don't give you
Most trend generators hand you one fixed clip: one length, one shape, one more reason to convert it afterwards. Here you set three things before the render starts.
- Length — 15 seconds, the full chase including the flip and getaway.
- Shape — vertical or horizontal, rendered that way from the start. No blurred bars bolted on after.
- Sound — generated with the picture in one pass, so the beat and the cut belong to each other. Nothing to sync in an editor.
And it's pay-once: credit packs bought one at a time. No subscription, nothing recurring, credits don't expire. A 15-second 768P clip is 50 credits. Full table on pricing.
Which photos work
The model copies the face, hair and clothes from each photo and keeps them consistent for the whole chase.
- Face the camera. Front-on or slightly turned. Profiles and long shots give it too little to copy.
- One person per slot. Crop out friends and posters with faces, or it may pick up the wrong one.
- Show the outfit. A waist-up or full-body shot keeps your clothes through the flip.
- Nothing over the face. No sunglasses, masks or hands. Caps and everyday glasses are fine.
- Real people who agreed. Yourself or someone who said yes. Celebrities, cartoon characters and minors are refused.
- Clear beats tiny. A sharp phone photo is enough. Blurry stills, heavy filters and tight head crops fail more often — fast motion already softens a few frames.
Length and shape: picking right
| Where you'll post | Length | Shape |
|---|---|---|
| TikTok / Reels / Shorts | 15 seconds | Vertical 9:16 |
| YouTube / X | 15 seconds | Horizontal 16:9 |
| First try | 15 seconds | Either |
The joke is the held distance, so it needs the squat, the close-up, the flip and the getaway. The studio always renders 15 seconds. Trim in the app if you want a shorter hook — trimming is free, re-rendering isn't.
The sound
H3 renders the audio with the video, so engine and road noise land on the picture without any editing.
For the trend sound, don't rip the audio out of someone's clip — add it inside TikTok instead: upload your video, tap Add sound, search for the trend sound, and turn the original audio down. That links your post to the sound page, where people browsing the trend find you. The story behind the sound is on the trend page.
What it costs
No subscription, nothing recurring. Buy credit packs one at a time: what you see is what you pay, and credits don't expire. You're only charged when a video finishes; failed renders go back to your balance automatically. Two photos are required on every run. A 15-second clip at 768P is 50 credits.
Common problems and fixes
| Problem | Cause and fix |
|---|---|
| Face doesn't match the photo | Use a clearer, front-on still with even light. Small, profile, or heavily filtered selfies are the usual cause. Fast motion always softens a few frames. |
| Render fails and mentions the photo | Safety check blocked the still — often because it looks like a public figure, or the frame has text or a logo. Credits return to your balance; swap the photo and try again. |
| Two people fighting in the frame | One person per photo. Faces and posters in the background confuse the reference — crop them out. |
| Slower than usual | Peak hours queue. The job runs to the end even if you close the page; the clip lands in Generation history. |
| I only want vertical | Pick 9:16 in step 5. Don't add black bars after — the scene is composed for that shape. |
FAQ
Can I make it for free? A 15-second MiniMax H3 clip at 768P is 50 credits. Packs are on pricing.
Do I need the original clip or a template? No. There's nothing to download. Pick a background. The chase prompt is sent in the background with your two photos.
Can two people be in it? Yes — Photo 1 is the rider, Photo 2 is the chaser, and both faces stay consistent through the whole clip. Both photos are required.
How long is the finished video? 15 seconds, 24fps, up to 2K, MP4, sound already in it.
Can I post it vertical? Yes, and it's rendered vertical from the start rather than padded afterwards. Pick the shape at step 5.
What makes it look close to the original? The shot structure is the trend's: the low squat with fists forward on handlebars that aren't there, the close-up cut, the mid-air flip, the getaway. Keep that framing and pick 15 seconds for the full beat. See the shot-by-shot breakdown on the prompt page.
How is this different from the other generators? Shape and sound are yours to set — 15 seconds, vertical or horizontal as rendered, audio generated with the picture — and you pay per pack instead of subscribing. MiniMax H3 generates from your two photos, the empty-road still, and a fixed prompt.