How to Make the Viral Carpool Karaoke AI Video

The Carpool Karaoke AI video format is a four-person in-car singalong: fixed seats, windshield-style cuts, and Walk It Talk It–style energy. On this site you can make one in a few minutes with four seat photos, a built-in ~15-second reference, 720p output, then preview and download.

Start here: Homepage generator

Note: Carpool Karaoke AI is an independent tool inspired by the trend — not an official Carpool Karaoke, Migos, or Late Late Show product.

Easiest method (this site)

1. Open the homepage generator and sign in.
2. Upload four clear photos — one per seat (see seat order below).
3. Choose 9:16 (TikTok, Reels, Shorts) or 16:9. Output is 720p and about 15 seconds.
4. Tap Generate — you get a clip with our built-in Carpool Karaoke motion reference (no prompt box on the homepage).
5. Preview the full clip, then download MP4. Add permitted music in TikTok / Instagram / CapCut if you need a specific track.

You do not need to upload the original reference performance video or paste a long Wan prompt when using this site.

Read this first: seat order is the casting plan

Upload order decides who sits where. On this site the mapping is:

| Slot | Seat | Tip | | ----------- | ------------------- | ------------------------------------------------------------ | | Image 1 | Front passenger | Clear face; sits shotgun | | Image 2 | Driver | Often the most expressive beat — put someone who will commit | | Image 3 | Rear left | One person only — no group crop | | Image 4 | Rear right | One person only — no group crop |

If you only have three people, duplicate someone into a rear seat instead of leaving a slot empty.

Why the reference motion matters

Tutorials that only use generic image-to-video plus a text description often miss the trend’s camera cuts and seat timing. What makes the meme recognizable is following a specific in-car reference. Here that reference is already applied — upload four photos and generate.

Four seats in one car is also why separate face-swap stacks look “assembled”: each pass lights faces differently. A single generation keeps one lighting and timing context across the cabin.

What you need before you start

• Four images — clear face (or upper body) photos; JPEG/PNG/WebP as the uploader allows
• Account & credits — sign in; when signup credits are enabled, new accounts receive trial credits
• Consent — only upload people you have rights to use
• Expectations — results vary; audio is not guaranteed; review every clip before posting

Step 1: Pick strong photos

Do

• Front or three-quarter view, eyes visible, minimal motion blur
• Even lighting; avoid heavy beauty filters
• One subject per file — head-and-shoulders works well

Avoid

• Group shots in a single slot (two faces confuse casting)
• Sunglasses or masks covering key features
• Extreme crops that cut off the jaw or forehead

More casting notes: Carpool Karaoke prompt guide.

Step 2: Open the generator

1. Open the homepage and go to Generate (/#generator).
2. Sign in if prompted.
3. Upload four seat photos in the labeled order.

You do not paste a prompt on the homepage; the Wan reference and seat mapping are applied for you.

Step 3: Choose format

| Setting | When to use | | -------- | ---------------------------------------------- | | 9:16 | TikTok, Reels, Shorts, Stories | | 16:9 | YouTube, X, desktop feeds | | 720p | Fixed output quality on the homepage generator | | ~15s | Fixed length to match the motion reference |

Step 4: Generate and preview

1. Check the credit cost on the Generate button.
2. Tap Generate and wait (queue and load vary).
3. Watch the full preview — check identity, seats, and hand/face glitches.
4. If you like it, download the MP4. If not, swap photos or retry.

Buy more credits from the result card or Pricing.

Step 5: Share responsibly

• Disclose AI when platforms or laws require it.
• Do not claim the clip is official Carpool Karaoke / Migos / Late Late Show content.
• Music: exports may not include cleared commercial audio; add licensed tracks in an editor if needed.

Advanced path (other tools)

If you build this yourself in another Wan / reference-to-video stack:

1. Prepare one composed still or four identity refs that match seat roles.
2. Drive motion from a fixed in-car performance clip.
3. Keep seats fixed — no character seat swaps mid-clip.
4. Export vertical 9:16 for short-form apps.

Example Wan-style seat prompt used conceptually on this site (homepage applies it for you):

Carpool Karaoke: four people singing in a car. Match the reference video’s camera, seats, timing, and motion. Image1=front passenger, Image2=driver, Image3=rear-left, Image4=rear-right. Keep each face identity, natural lip sync. No extra people, no text, no watermark.

Copyable variants live on the prompt guide.

Troubleshooting

| Problem | What to try | | ----------------------- | ---------------------------------------------------- | | Wrong person in a seat | Re-check Image 1–4 order | | Face looks wrong | Brighter photo, less crop, one face only | | Empty / weird rear seat | Don’t leave slots blank — duplicate a photo | | Not enough credits | Pricing or sign-in trial if enabled | | Want a different scene | Image to video for custom prompts |

Quick checklist

□ Four clear single-subject photos
□ Seat order: front passenger → driver → rear L → rear R
□ Signed in with enough credits
□ 9:16 or 16:9 chosen
□ Full preview watched
□ Consent + AI label planned for upload

Ready? Make your Carpool Karaoke AI video.