GoKawiilDeveloper builds differentiable-rendering tool to auto-arrange 3D meshes into anamorphic sculptures
A developer inspired by Jonty Hurwitz's anamorphic sculptures built a custom application that arranges arbitrary 3D meshes—like cars, pigs, and pigeons—so their combined silhouette forms a target image when viewed from one specific camera angle. The system starts with randomly placed objects and uses gradient descent with a differentiable renderer to iteratively adjust each mesh's position, size, and rotation until the projected outline matches the target picture.
GoKawiil's interpretation of the reporting above, not reported fact.
This project shows how differentiable rendering, a technique increasingly used in computer graphics and machine learning, can be applied to solve creative geometric puzzles once reserved for skilled sculptors doing manual trial and error. It highlights a growing trend of using gradient-based optimization to automate complex spatial arrangement problems that traditionally required artistic intuition and physical prototyping.
- The tool arranges arbitrary 3D meshes into a specific silhouette visible only from one camera angle, mimicking anamorphic art.
- Differentiable rendering is essential, allowing errors between the rendered silhouette and target image to backpropagate into mesh position, scale, and rotation.
- A carefully tuned loss function is required so the optimization captures fine details rather than settling for a roughly correct overall shape.
Source: tncardoso.com, 2026-09-11
Published there as: “Optimizing Anamorphic Sculptures”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.