Your Kid’s Stick Figure Can Actually Run! This Meta Open Source Project Kept Me Busy All Day

I have a young nephew who loves drawing. Every time he finishes a knight with a sword or a dancing figure, he asks me: “Uncle, can it move?”

I used to brush it off with “you can learn animation when you grow up.” Until I stumbled across facebookresearch/AnimatedDrawings—an open-source project from Meta that turns children’s sketches into real animations. I spent an afternoon playing with it, and now I’m writing this article.

What Does This Project Actually Do

AnimatedDrawings stands for “A Method for Animating Children’s Drawings of the Human Figure.” Very academic name, very magical results.

You take a photo of a child’s stick-figure drawing, upload it to the system, and it automatically identifies the head, torso, arms, and legs. Then it rigs a skeleton onto it. After that, you pick a preset action—walking, running, jumping, dancing—and the stick figure comes to life, moving convincingly.

Not some crude stretch-and-distort effect. Real skeletal animation. Arms swing, legs bend, body bobs up and down. All of it.

The Core Tech: Not Just Simple Image Warping

At first I thought this was just ordinary image warping or GIF generation. After reading the paper and code, I realized it’s much more sophisticated.

Extracting Characters from Drawings

Step one is character extraction. The system uses computer vision models to separate the drawn figure from the background. This step handles very real-world problems—kids rarely draw their figure neatly in the center of a blank page. There’s usually a sun, a house, some flowers nearby. The system has to be smart enough to grab only the character.

Automatic Skeleton Rigging

Step two is what I find most amazing. It doesn’t require you to manually label arms and legs. The model looks at the drawing and infers the body’s topology, then automatically generates a skeleton rig. Even if the child drew one leg longer than the other, or arms growing out of the chest, it finds a way to adapt.

It uses a clever approach: instead of forcing a standard human skeleton onto the drawing, it flexibly deforms based on the actual proportions of the sketch. So each person’s drawing moves with a unique “drawing style” rather than becoming a cookie-cutter template.

Motion Retargeting

Step three redirects preset motion capture data onto this custom skeleton. Meta used lots of real mocap data—walking, running, cartwheels, you name it. The system “translates” these motions to fit the proportions and style of the drawing.

The result is a crooked stick figure taking steps as fluidly as a professional animated character.

Real Experience: From Photo to Animation in Five Minutes

I tried the full pipeline with one of my nephew’s drawings. The steps were roughly:

git clone https://github.com/facebookresearch/AnimatedDrawings.git
cd AnimatedDrawings
pip install -e .

Then I scanned the drawing into a PNG and ran the project’s built-in script:

from animated_drawings import render
render.start('./examples/config/mvc/child_drawing.yaml')

About two minutes later, an MP4 popped out. The little figure my nephew drew was actually walking across the screen. His eyes went wide.

The whole workflow isn’t too complex for developers, but probably still has some barriers for average parents. The project provides several preset examples—just tweak the config files and you’re good to go.

Pros and Cons

On the plus side: the results are genuinely stunning, especially the automation level of skeleton rigging; the code is open source so researchers can build on it; the preset motion library is quite rich, covering most common movement types.

The downsides are real too: it wants decent hardware, GPU acceleration preferred, pure CPU is a bit slow; it currently only supports roughly humanoid characters, so if you draw a cat or a car, it won’t know what to do; also, while motions are based on real mocap data, retargeting onto stick figures occasionally causes clipping or joint twisting that needs manual tweaking.

More Than a Toy

Most people’s first reaction to something like this is “cool.” But the more I thought about it, the more I felt it has deeper significance.

Children’s drawings are direct outputs of their imagination. Before, they stayed on paper. Now they can come alive. That instant feedback is huge motivation for a child’s creative drive. After watching his drawing move, my nephew immediately drew three more and said he wanted to make a “movie.”

From a technical angle, this project also shows a trend: AI and computer vision are moving toward “understanding non-standard inputs.” Children’s drawings have no perspective, messed-up proportions, and messy lines—traditional algorithms couldn’t handle them at all. But modern models have learned to “understand” these drawings like humans do, rather than mechanically demanding standard inputs.

How It Compares

There are commercial apps that can animate drawings too, especially in some children’s educational software. But they’re mostly closed systems—you can only use their built-in templates, no customization. AnimatedDrawings’ advantage is being fully open source. You can tweak model parameters, add new motions, even train your own skeleton rigging strategies.

If you just want to bring your kid’s drawings to life, commercial apps might be more convenient. But if you want to study the tech behind it or build secondary applications, AnimatedDrawings is pretty much the only choice.

Who Should Use It

Parents looking to surprise their kids, researchers exploring non-standard image animation, developers wanting to integrate into educational apps—any of these groups should give it a try. If you just want quick results, the project’s example configs are sufficient. If you want deep customization, the code comments are reasonably detailed.

I’m planning to spend next weekend drawing a whole batch with my nephew and making a complete “animated film.” He draws, I make them move. This might be the most fun parent-child project of the year.

---

About the Author

Liudingyu is a full-stack developer and heavy GitHub user. With 900+ starred repos over the past 3 years, this site only covers tools I’ve actually used or deeply researched.

📧 Found a great tool to recommend? Email [email protected]