Attic Dragon

1. Context

The project began as a lighting and look-development image built around a Chinese incense burner in an attic. I used the scene as a practical test bed for a Houdini Digital Asset that could generate both organised spiderwebs and older, damaged cobwebs.

2. Production problem

The scene needed many fine strands connecting irregular surfaces. Placing every curve by hand would make iteration slow, while a fully automatic result would give too little control over composition, density and visibility in the final shot.

Early geometric patterns also felt too regular. The tool needed to retain a readable web structure while introducing the sag, breakup and unevenness expected from an old attic.

3. My role

I designed the procedural workflow, implemented the VEX-based generation, packaged the solver stages as Houdini Digital Assets and exposed artist-facing controls. I also built the test scene, developed the web material, lit the shot and integrated the generated curves into the final image.

4. Solution

The workflow separates artist input, preprocessing, generation and final simulation. This keeps placement decisions visible while allowing the underlying procedural stages to be reused.

1. 01 Artist input Scene geometry, anchors and painted mask
2. 02 Preprocess Prepare eligible regions and proximity data
3. 03 Solver choice Spiderweb, cobweb or scene-cobweb variant
4. 04 Simulation Process the generated curves
5. 05 Shot integration Lookdev, lighting and final composition

Artist-facing inputs

• Scene geometry used to evaluate possible connections
• Painted masks to restrict generation to useful regions
• User-defined anchor points for deliberate placement
• Size, shape and density controls for shot-specific variation
• Curve shaping for looser or more tensioned results
• Separate structured-web and deteriorated-cobweb modes

Without web / with generated web

Generated results and final integration

5. Production constraints

Topology and placement

The documented workflow starts from scene geometry, user anchors and a painted mask. Proximity and object shape guide eligible regions; the archived breakdown does not document behavior for non-manifold or otherwise invalid input meshes.

Density and tension

Density, size and shape remain artist-facing decisions. Curve shaping was used to move away from rigid straight connections and create looser, more organic tension.

Generation time

No measured cook time was preserved for the published build, so this page does not claim a generation-time figure. A repeatable timing test still needs to be recorded.

Rendering thin geometry

The strands had to read in the shot without becoming opaque cables. Translucency, specular response and lighting were adjusted together during look development.

Artist control

Masks and anchor points keep the procedural result subordinate to composition instead of scattering detail indiscriminately across the scene.

Packaging and maintenance

The distributed workflow is divided into preprocessing, web, cobweb, scene-cobweb and simulation HDAs. The internal networks remain packaged for distribution, while this page documents the modules, artist inputs and public system boundaries.

Problems encountered and procedural choices

6. Results

The HDA produced structured and deteriorated web variants from artist-defined regions and made their placement visible before final rendering. The same workflow was used in the Attic Dragon scene, connecting procedural tool development directly to a finished lighting image.

No defensible timing or error-reduction metric was recorded for this personal project. The evidence shown here is therefore visual and architectural rather than presented as a production benchmark.