DEV Community

Jordt Catena
Jordt Catena

Posted on

Mexico Pterodactyl AI Video: A Photo-Prep Tutorial (With a Tiny Python Pre-Flight Check)

If you have been anywhere near TikTok this month, you have seen the Juárez pterodactyl. First it was a blurry security-camera clip from Ciudad Juárez, then a fact-check (Snopes rated one widely shared clip "Fake" on November 17), and then the jokes: people "walking" the pterodactyl on a chain, or keeping it on the roof like a pet.

One tool built for this trend, Mexico Pterodactyl AI, is a template tool: you pick one of two clips, upload one photo, and press a button. There is no prompt and no timeline. That sounds like there is nothing to tune, but there is one variable you fully control, and it matters more than anything else: the photo.

This post is about that one variable. It is a photo-prep tutorial for developers who would rather get it right on the first render than burn credits guessing.

Why the photo is the whole game

The site is explicit about what changes and what stays. In the template, the pterodactyl, the location, the camera moves and the sound are fixed. What comes from your upload is the person: face, hair and the clothes you are wearing in the photo.

So think of it like a function with one argument:

video = template(person_from(photo), details_text?)
Enter fullscreen mode Exit fullscreen mode

Everything downstream depends on how much clean signal photo carries about your face and outfit. Garbage in, uncanny valley out.

The rules the site gives you

From the homepage, a photo that works has:

  • One person in it (group photos are on the "avoid" list)
  • A sharp face, turned to the camera
  • Whole body or waist up
  • Good light on the face (face a window or step into daylight)
  • The clothes you want to wear in the video
  • No beauty filter — use the photo straight from the camera

And things to avoid: sunglasses or a hand over the face, heavy filters, and a tiny face far in the background.

That is a pretty clear spec. The rest of this post turns it into a repeatable routine.

Step 1: Shoot (or pick) for the template you want

There are two video templates:

  • On a chain: one handheld shot of about 10 seconds with no cuts. You walk the pterodactyl across a shiny black-tiled hall, the camera swings to its beak and collar, then comes back to you sorting out the chain.
  • Rooftop pet: you stand next to your "pet" pterodactyl on a rooftop.

Because the original On a chain clip features a man in a black hat, a hat and boots in your photo go a long way. That is the site's own tip, and it is a good one: wardrobe that matches the vibe of the source clip makes the swap feel less pasted-on.

Step 2: Light the face, not the room

The easiest way to ruin the result is a backlit face: a window behind them, a dark silhouette in front. Turn around. Face the light source. If you are indoors, stand a step back from a window and let it hit you straight on.

Step 3: Kill the filters

Beauty filters smooth skin and reshape features. The model then has to reproduce a face that does not quite exist. Use the camera's raw output (or a normal JPEG from your phone's default camera app) instead of a social-app capture.

Step 4: Run a pre-flight check

Rather than eyeballing every photo, I wrote a tiny pre-flight script with Pillow. It does not know anything about the site's internals (and the site does not publish any upload limits that I could find), so it only checks the general stuff that the photo guidance implies: is it portrait-ish, is it big enough that the face will not be tiny, and is it bright enough.

from PIL import Image, ImageStat

def preflight(path, min_short_side=720, min_brightness=90):
    img = Image.open(path).convert("RGB")
    w, h = img.size
    issues = []

    if min(w, h) < min_short_side:
        issues.append(f"small image ({w}x{h}); your face may end up tiny")
    if w > h:
        issues.append("landscape photo; waist-up or full-body portraits are easier")

    # average brightness of the centre third, where a face usually is
    box = (w // 3, h // 6, 2 * w // 3, h // 2)
    brightness = sum(ImageStat.Stat(img.crop(box).convert("L")).mean)
    if brightness < min_brightness:
        issues.append(f"centre looks dark (mean {brightness:.0f}); face the light")

    return issues or ["looks fine, go render"]

if __name__ == "__main__":
    import sys
    for line in preflight(sys.argv[1]):
        print("-", line)
Enter fullscreen mode Exit fullscreen mode

The thresholds are my own guesses, not anything the site specifies. Tune them for your phone. The point is to catch the obvious misses (a dark, distant, landscape snapshot) before they cost you a render.

Step 5: Use "Add your own details" sparingly

There is an optional text box called Add your own details. It is for small things about you: a hat, a leather jacket, a chain, a watch. It does not change the scene, the pterodactyl, or the moves. Writing "on a beach at sunset" will not move the hall to a beach; the template still decides the place.

Keep it to one short line, for example in a white cowboy hat.

Step 6: Render and walk away

Tick "I have the right to use this photo" and press Make my pterodactyl video. Most videos take 3 to 6 minutes. You do not have to babysit the tab; the result lands in "My photos & videos" and you get an email.

The output is a vertical 9:16 clip with the template's sound, rendered at 480p (Standard) or 720p (HD) depending on the pack. The site says there is no watermark from them, and if a render fails on their side, it goes back to your account automatically.

Before you post

Two things I would not skip:

  1. Turn on the platform's AI-generated label. The site asks you to, and given that the whole trend started with people arguing over whether a clip was real, it is the honest move.
  2. Only use photos you have the right to use. The terms forbid photos of public figures or anyone who has not agreed. Your tío can be on the chain only if he is in on the joke.

Wrap-up

Template tools move all the creative decisions into one input. If you treat that input like a spec (one person, sharp front-facing face, waist up or full body, real light, no filter), you get a clip that actually looks like you instead of a vaguely related stranger walking a dinosaur.

Top comments (0)