A melting ice cube simulation in Houdini for beverage CGI is most reliable when the ice deformation, liquid runoff, droplets, and condensation are treated as connected but separate systems. For most advertising shots, the strongest approach is a hybrid: use procedural or VDB-based deformation to control the cube’s silhouette and volume loss, then source only the visible meltwater into a focused FLIP simulation.
This workflow preserves art direction while allowing rivulets, contact sheets, and puddles to respond naturally to the glass and supporting surfaces. The process below covers strategy selection, simulation-ready geometry, SDF handling, FLIP sourcing, collision design, meshing, caching, retiming, and the refractive materials required for a polished beverage render. It also explains where physically plausible behavior should give way to deliberate control of timing, runoff paths, and product readability.
Choose the melting strategy from the visible action
“Melting ice” is not one effect. The cube may lose volume, its corners may soften, a wet film may spread over the surface, water may run toward the glass, droplets may detach, and a puddle may form below. Condensation on the glass is another process again. Asking one solver to create every layer makes the shot expensive and usually reduces control.
Start by identifying what the camera must resolve. A medium product shot may need only a changing silhouette, wet highlights, and two or three designed rivulets. A macro shot focused on the coaster may require physically credible pooling, adhesion, droplet breakup, and contact with the glass. Shot duration matters as well: a subtle two-second melt can be built very differently from a long time-lapse transition.
Procedural deformation
Use animated geometry when the principal requirement is a controlled silhouette. Point deformation, localized shrinking, remeshing, and rest-space masks can soften corners and create uneven recession without the overhead of a fluid solve. Separate curves, meshes, or instanced droplets can provide the visible water.
This method is appropriate when the cube remains mostly intact, liquid movement is secondary, and timing must be revised quickly across product variants. It should not look like a uniform scale-down. Retain mass in protected areas, erode contact regions more aggressively, and expose different faces at different rates.
Volume-based melting
Use an SDF or VDB workflow when the cube must visibly lose material or develop cavities while remaining watertight. Convert the prepared ice to a signed distance field, modify the field with animated erosion masks or subtraction volumes, and polygonize it after the large-scale shape is approved.
Volume processing produces coherent topology, but it introduces resolution limits. A feature narrower than a few voxels may vanish, flicker, or become disconnected. Keep broad erosion in the volume branch and reserve bubbles, microcracks, frost, and crystalline texture for interior geometry, displacement, or shading.
FLIP and the hybrid workflow
FLIP becomes useful when water must discover a path around the cube, cling to the glass, bridge contact points, split into rivulets, or accumulate on a coaster. It solves liquid motion; it does not determine how a solid cube changes phase. The ice recession and the liquid source therefore need to be designed explicitly.
For most hero beverage shots, combine VDB or procedural ice deformation with a bounded FLIP solve for the visible runoff. This preserves control over the recognizable cube while giving the water credible collision and pooling behavior. Fine beads, thin films, and condensation can remain separate secondary layers rather than forcing the primary FLIP simulation to resolve every scale.
- Use procedural deformation for a designed silhouette, rapid iteration, and limited visible runoff.
- Use VDB erosion for coherent volume loss, cavities, and changing topology.
- Use FLIP when liquid motion, contact, breakup, or pooling is a focal element.
- Use a hybrid when both the ice shape and liquid path require independent art direction.
Prepare the ice and scene at a consistent scale
Simulation stability begins with scene scale, transforms, and geometry preparation. Establish Houdini units and the intended real-world dimensions before choosing voxel size, particle separation, surface tension, or absorption distance. Apply transforms, remove accidental non-uniform scale, and confirm normals and orientation before converting anything to volumes.
Build separate simulation and render assets
Start with an intentionally beveled cube. A perfectly sharp primitive edge looks synthetic and can create fragile regions when eroded. Use enough geometry to describe the major edge profile, then add restrained asymmetry so the corners do not appear cloned. Preserve the familiar cube proportions until the shot calls for more aggressive deformation.
Maintain separate simulation and render branches. The simulation mesh should be clean, connected, and no denser than required for deformation or volume conversion. The render asset can carry smoother bevels, displacement, interior bubbles, and crack geometry. Increasing solver density to capture refractive microdetail rarely improves the melt behavior.
Keep internal structure independent from the outer surface. Cloudy cores, bubbles, fractures, and density variation can be stored as volumes, points, interior meshes, or shader masks. That allows the surface to change without rebuilding all the internal detail and makes material revisions possible after the simulation is approved. For a broader treatment of these material cues, see photorealistic ice in beverage CGI.
Choose voxel size from the smallest important feature
Inspect the SDF rather than assuming a successful conversion. The grid must preserve the bevel, lower contact zone, and any cavity that affects the final silhouette. Features represented by only one or two voxels are likely to round off or change unpredictably between frames.
A practical production setup often uses more than one field resolution. Keep a moderate-resolution master SDF for broad deformation and source generation, then build a locally refined or higher-resolution render volume only where the camera needs it. This is easier to manage than carrying a hero-resolution grid through every iteration.
Prepare collision proxies by physical role
Create explicit collision inputs for the glass, beverage, coaster, tabletop, and any garnish or prop that can redirect water. These should represent the surfaces relevant to the solve rather than every decorative feature of the render mesh.
The glass proxy needs a consistent inner wall, outer wall, rim, and base if runoff interacts with them. Inspect the collision SDF for sign, wall thickness, open boundaries, and undersampled bevels. A thin or inverted field can produce leaks, trapped particles, or a visible gap between water and glass.
Treat the beverage surface according to the shot. If meltwater merely reaches an opaque or visually stable drink, a controlled collision or termination surface may be sufficient. If the water must enter, displace, or mix with the beverage, the liquids need a compatible shared simulation strategy; a static collider will not represent that interaction.
Use clear names and versions such as ice_sim, ice_render, glass_collision, and coaster_collision. Cache converted collision fields when they are reused. Before a full solve, run a short wedge that confirms the cube retains its silhouette, the colliders prevent penetration, and the smallest required features survive volume conversion.
Art-direct the changing ice volume
The phase-change look should be driven by a persistent control field rather than scattered animation values. Create a float attribute such as meltmask or melt on the simulation-ready ice and combine temporal progression with spatial influences: normalized height, curvature, surface exposure, contact proximity, and painted retention masks.
Keep low-frequency decisions separate from high-frequency breakup. First establish when the melt begins, which corner remains recognizable, how quickly the lower contact region recedes, and the amount of final volume loss. Add restrained noise only after those beats are approved. Fine noise introduced too early tends to make the cube look chemically dissolved rather than thermally softened.
Drive VDB erosion without losing the silhouette
Convert the original cube to an SDF and use the melt field to control erosion, offsetting, or smooth subtraction. In Houdini, this can be assembled with VDB reshape, combine, smooth, and resample operations or equivalent volume-processing nodes. The important point is to keep the control field and the ice SDF available separately.
A hybrid deformation branch can protect art-directed features after the broad volume loss. For example, use the VDB result to establish mass reduction, then blend or project selected regions toward a protected silhouette. This makes it possible to hold a camera-facing corner through the product beat without freezing the entire cube.
Temporal filtering is essential. Abrupt changes in an erosion mask cause popping geometry and discontinuous liquid emission. A SOP Solver can accumulate melt over time so the field increases progressively, while a separate animated multiplier controls the shot-level schedule. Keeping timing upstream also makes later retiming less destructive.
Preserve attributes through topology changes
Store useful data before remeshing or VDB conversion. Rest position, melt amount, surface age, retention masks, source IDs, and contact classifications can drive both sourcing and shading. Transfer them back to the polygonized surface using the original geometry or persistent volumes as references.
Keep the initial and current ice SDFs as distinct caches. Their difference provides a useful estimate of where material has receded. It is not automatically a physically exact phase-change model, particularly after filtering or art-directed deformation, but it supplies a coherent basis for locating and budgeting the meltwater source.
Source and simulate the meltwater
Emit water from meaningful recession zones rather than the entire wet-looking surface. Derive a source mask from the SDF difference, surface exposure, or accumulated melt attribute, then filter it spatially and temporally. A second art-direction mask can delay emission, protect branding areas, or favor runoff on the camera-facing side.
Do not assume that every removed voxel must become visible FLIP volume. SDF smoothing, unit conversions, and deliberate exaggeration can make strict one-to-one conversion impractical. Use the removed volume as a reference, then expose a source-rate control that preserves continuity and keeps the puddle within the desired scale.
Set source velocity deliberately
A melt source should not fire particles directly away from the cube. Project gravity or a guide direction onto the local surface tangent, then add only enough outward velocity to prevent immediate penetration. Transfer source normal, velocity, melt amount, age, and an identifier that distinguishes meltwater from other fluids.
Source filtering prevents minor VDB changes from creating bursts on every frame. Minimum source thickness, temporal averaging, and an emission threshold can suppress unstable patches. Vary timing and velocity subtly where a repeated stream would otherwise look mechanical, but avoid high-frequency randomness that breaks liquid continuity.
Resolve runoff and contact at the appropriate scale
Particle separation determines which liquid features the solve can represent. Select it from the narrowest hero rivulet or bead that must survive, then restrict the domain to the visible action. Use substeps when fast source motion or changing collisions allow particles to cross thin features between solver updates.
Adhesion, viscosity, and surface tension solve different problems. Adhesion influences how readily the water remains near the glass; viscosity changes resistance to deformation; surface tension helps small forms remain cohesive. Tune each against the scene scale and reference behavior rather than increasing all three to make the water feel “thicker.”
If gravity produces a plausible but compositionally weak path, guide the broad flow instead of constraining every particle. A shallow guide surface, directional velocity field, or painted source mask can steer the runoff toward a readable highlight or away from a label. The water should still respond to collisions and local curvature.
Separate liquid into production layers
- Primary runoff and pooling: Use FLIP for streams, contact sheets, and accumulated volume.
- Thin wet films: Use a surface layer, shader mask, or dedicated secondary mesh where a resolved liquid volume would be inefficient.
- Hero droplets: Generate secondary particles or instances from curvature, velocity, and contact masks.
- Condensation: Build a separate glass-based system controlled by cooling zones, composition, and branding clearance.
A puddle created only by unconstrained runoff may spread behind the product or beyond the useful frame. It is reasonable to guide the receiving area or build a separate shallow pooling pass, provided the contact line and incoming runoff remain consistent. Similar layering principles apply to ice, garnishes, and liquids in cocktail CGI for advertising.
Mesh, cache, and retime for production
Simulation particles are not render surfaces. Build the meltwater mesh in a separate branch by converting particles to an SDF, filtering the field, and converting it to polygons. Keep ice, runoff, puddle, droplets, and condensation separable so each can be revised or rendered independently.
Reconstruct the liquid without erasing rivulets
Voxel size is the main meshing decision. A coarse field is stable and economical but can close narrow channels and round contact lines. A fine field preserves rivulets but increases memory, polygon count, and sensitivity to gaps in particle coverage.
Apply moderate filtering in SDF space before polygon conversion. Volume filtering generally produces a more coherent result than heavy polygon smoothing, but too much will remove the details that explain the direction of the melt. If polygon cleanup is still required, project the smoothed result back toward the reconstructed surface to limit volume drift.
Judge thin features through the delivery camera. A physically present film may disappear under antialiasing or motion blur, while a tiny droplet can become an oversized refractive highlight. Preserve the forms that communicate attachment and direction. Replace subpixel simulation debris with controlled instances or omit it.
Protect contact points and topology
Inspect the ice mesh for holes, self-intersections, non-manifold regions, flipped normals, and detached fragments. Recalculating normals cannot repair overlapping shells or invalid topology. Clean those defects before evaluating a refractive shader, because they can appear as black patches, broken transmission, or discontinuous interior shading.
Use the approved collision fields as references when projecting or masking liquid near the glass and support surfaces. A slight controlled overlap is generally less distracting than a visible gap, but excessive penetration makes water look embedded in the glass. Check contact from the render camera and in motion.
Cache by dependency
Cache the changing ice, control fields, source volumes, collision SDFs, FLIP particles, reconstructed liquid, and secondary droplets at separate stages. Record the scene scale, frame range, substeps, particle separation, voxel size, source version, collision version, and meshing settings with each cache.
This separation avoids rebuilding the whole shot when only the puddle footprint or mesh filtering changes. It also makes diagnosis possible: source attributes reveal whether a failed rivulet began in the melt mask, source conversion, collision setup, or fluid solve.
Approve timing before expensive secondary work. Retime the driving melt progression or primary cached motion first, then regenerate dependent emission, droplets, and motion vectors. Interpolating a changing-topology liquid mesh can produce collapsing strands and unstable contact; particle or volume caches usually provide safer retiming inputs than the final polygons.
Shade and light the refractive layers separately
Ice, meltwater, beverage, and glass should not share one generic transparent material. Their indices of refraction may be close enough to create subtle interfaces, but their absorption, roughness, internal structure, thickness, and visual roles differ. Separate materials and geometry branches also make compositing corrections possible without resimulation.
Build a volumetric ice response
Clear transmission alone makes ice resemble acrylic. Add restrained variation in absorption, scattering, roughness, and internal density. Cloudy regions, bubbles, fractures, and granular inclusions should establish depth without turning the cube into an opaque white block.
Use broad features that survive refraction and the final camera distance. Thickness or an approximation of travel distance can modulate absorption, while crack and density masks can influence scattering and roughness. Preserve clearer paths through the body so the beverage color remains visible.
Wet ice needs localized treatment. A proximity or contact mask can reduce roughness and introduce a continuous water layer around active melt zones. Avoid making the entire cube uniformly glossy; the contrast between frosted, melting, and wet regions helps communicate the phase change.
Differentiate water, beverage, and glass
Meltwater should emphasize thickness, reflection, edge highlights, and contact. Thin streams often need controlled environmental reflections to remain visible, while puddles need a coherent interface with the coaster or tabletop. Inconsistent normals or an excessively thin closed surface can make water render as dark geometry.
The beverage requires its own absorption and scattering response so its color stays stable through the ice and glass. Glass needs physically coherent thickness and transmission, but render geometry may require cleanup where multiple refractive shells overlap. Excessive absorption or unnecessary nested surfaces can make the product muddy and increase noise.
Keep condensation as a separate geometric or shading layer. Vary droplet size, spacing, and density with masks rather than uniform procedural noise. Reduce or remove droplets where they obscure branding, key highlights, or the beverage color.
Light interfaces rather than transparency alone
Large controlled sources or reflection cards can define the ice silhouette, glass rim, and runoff trails. A background with deliberate tonal transitions reveals refraction and thickness more effectively than a flat value. Use restrained fill to preserve the drink’s color without flattening the specular structure.
Test motion blur at delivery resolution. Narrow streams can become faint streaks or disappear when their width approaches a pixel. After retiming, revisit shutter duration, deformation blur, velocity blur, and sampling rather than assuming the original settings remain valid.
Useful AOVs include masks for ice, meltwater, beverage, glass, and condensation, along with reflection, refraction, shadow, contact, depth, position, normal, and object or material IDs. These passes help distinguish a material problem from a mesh, lighting, or compositing problem.
Diagnose common failure modes in the correct stage
The cube shrinks uniformly
The melt field is too simple or has been reduced to a global scale value. Separate onset, regional rate, corner retention, and surface exposure. Use rest-space or volume masks so selected areas recede at different rates, and keep liquid release independent from visible ice deformation.
Water leaks through or floats above the glass
Display the collision SDF as the solver sees it. Check scale, sign, thickness, rim resolution, concave regions, transforms, and the relationship between simulation and render geometry. Increase substeps for temporal tunnelling and collision resolution for spatial gaps; they are not interchangeable fixes.
Rivulets disappear during meshing
Review particle coverage, voxel size, SDF threshold, filtering, and minimum-component removal. Preserve important streams in a dedicated reconstruction branch or local high-resolution pass rather than raising the resolution of the entire solve.
The liquid surface flickers
Look for unstable particle support, isolated components, inconsistent VDB filtering, changing reconstruction settings, or aggressive topology cleanup. Cache the reconstructed field once the meshing parameters are approved. Test without motion blur to separate geometric instability from render sampling.
The ice looks rubbery, waxy, or plastic
Uniform roughness and a clear surface are usually the main causes. Restore edge definition, thickness-dependent absorption, subtle internal inclusions, and localized wetness. Random surface noise is not a substitute for believable internal structure and often weakens the designed silhouette.
The render becomes dark or noisy
Inspect overlapping refractive layers, shell thickness, normals, absorption distance, refraction depth, and sample settings. Render ice, water, beverage, and glass separately to identify the failing interface. Simplifying an invisible nested surface can be more effective than increasing samples indiscriminately.
The shot is too expensive to revise
Return to dependency-based caches and camera-based detail. Use proxy colliders and moderate-resolution solves to approve silhouette, timing, and runoff direction. Reserve high-resolution VDB conversion, local FLIP detail, droplets, motion blur, and refractive sampling for regions that affect the final frame.
Expose shot-level controls for melt onset, regional rate, silhouette retention, source strength, runoff direction, puddle footprint, droplet density, and condensation coverage. When those controls remain upstream of clearly versioned caches, the Houdini melting ice simulation can be revised without sacrificing the deliberate composition expected in beverage advertising.