The Hug

We started by taking hundreds of photographs of two people hugging. The models had to hold their poses completely still because any movement would make them look deformed when turned into a 3D model.

For this piece we needed to scan both people separately, so we carefully set up tripods to support our limbs while the other person climbed out of the embrace. One less technical side note: it was important to work with two people who actually wanted to be hugging. If the pure human emotion isn't present in the source, no amount of digital wizardry can put it there.

Switching to video

We switched to capturing video instead of photos so that the hugger could be captured quicker and their inevitable movement would be less problematic. Video has less detail than photos, but the trade-off between resolution and movement made the final 3D model more accurate despite using a lower quality data set.

Polycam and Rhino

We uploaded the videos to Polycam, and when the mesh was ready we decimated it so that it had fewer triangles and would be easier to work with in Rhino.

Blender cleanup

We manually cleaned the meshes in Blender to remove the tripods and close the holes, as well as align the models because they had shifted slightly during capture. Then we exported low-poly versions and brought them into Rhino and Grasshopper.

Iteration 1: The invisible hugger

For our first experiment, we subtracted one hugger's body from the other's, so that you could see the imprint that a hug leaves on them. The invisible hugger. This ended up looking like not much, and was a ton of work, and totally failed to convey the emotion we wanted.

Iteration 2: Stretching the vertices

For our second experiment, we stretched the models' legs so that they would look delicate and precarious in their hug, by translating each vertex in the mesh downwards exponentially according to its distance from a parametrically set midpoint threshold — you get to choose where to start stretching them. This didn't convey the emotion either. The vertex translation script let us control the stretch threshold parametrically, with real-time feedback in the Rhino viewport.

Iteration 3: Ginkgo leaves

For our final experiment we deconstructed the mesh to get a list of each face on it, as well as its normal — which direction the face is facing. We drew a NURBS curve based on a ginkgo leaf we found under a tree, and deformed it to match the curve of our real leaf. Then we made a copy of the leaf for every face on the model's mesh. This ended up crashing Grasshopper because there were 2,000 leaves, so we added a sparseness control and randomized each leaf's Z direction so that it looked more organic.

Finally we created a script to generate planes for the leaves blowing off the mesh. We take in the original mesh, use its maximum and minimum values along with a wind direction and wind force, and randomly scatter new leaves flying off the figure. We also multiplied the x and y values of the leaves by their own exponent to cluster the leaves around the model — then got rid of half the leaves because it just felt right.