Beverage Splash CGI in Houdini: Hero Liquid Shot Workflow

A production-ready Beverage Splash CGI in Houdini workflow begins with the final image, not the FLIP Solver. The camera, product silhouette, impact frame, liquid scale, shutter, and delivery resolution determine which details must be simulated and which can be added later. Once those constraints are fixed, the work can be divided into stable collision and source preparation, a controlled Houdini FLIP fluid simulation, camera-aware meshing, beverage shading, and render validation.

This workflow explains how to preserve readable crowns, sheets, tendrils, and droplets without letting secondary spray overwhelm the product. It also covers the production decisions that often determine whether a promising simulation becomes a usable hero liquid shot: wedge testing, cache versioning, retiming, diagnostic renders, and independent handoff of simulation, mesh, and look-development data.

Plan the hero splash before building the FLIP network

A beverage splash is a designed product image supported by fluid dynamics. Real-world reference is essential for understanding momentum, sheet thickness, crown formation, droplet scale, breakup timing, and contact behavior, but it should not become an inflexible specification. A physically accurate impact can be too brief, low, or chaotic for advertising composition. The objective is physically plausible continuity with an art-directed silhouette and timing.

Start by identifying the central event: a pour hitting a filled glass, a bottle entering liquid, an ice cube impact, a carbonation burst, or a suspended splash wrapped around packaging. Each action implies a different source, collision arrangement, and balance between connected liquid and secondary droplets. For a related treatment of continuous streams and impact behavior, see this guide to photorealistic liquid pouring CGI.

Define the editorial beats

Mark the frames on which the action must read:

1. Approach: the source or collider establishes direction and speed.
2. Initial contact: the point of impact is visible and credible.
3. Maximum pose: the crown or main sheet reaches its strongest silhouette.
4. Breakup: tendrils and droplets separate without obscuring the product.
5. Settle or exit: the liquid resolves cleanly within the edit.

These beats are more useful than simulating a long, unconstrained sequence. They provide acceptance criteria for low-resolution wedges and make it clear whether a revision concerns timing, energy, silhouette, or surface detail.

Establish a visual hierarchy as well. The connected liquid body should explain the event. Medium breakup pieces extend its direction, while isolated droplets provide scale and energy. Mist, foam, bubbles, and micro-spray are tertiary elements. If every particle receives equal visual emphasis, the splash becomes a field of highlights rather than a product-led composition.

Lock scale, camera, and delivery conditions

Build at consistent real-world scale before tuning gravity, viscosity, surface tension, collision thickness, or particle separation. Rescaling the scene after simulation is not equivalent to resizing a static model: it changes the relationship between fluid properties, velocities, time integration, and collision resolution.

Approve the camera early. A close, backlit sheet exposes holes, unstable edges, and surface noise that a wider packshot may hide. Conversely, a wide shot may not benefit from expensive micro-droplets but still requires a strong silhouette and clean product contact. Judge detail by its projected size in the final camera rather than by how impressive it looks in a free viewport.

Record the delivery resolution, frame rate, shutter, focus treatment, and likely retiming range. Thin features that survive an unblurred viewport preview may disappear after motion blur, depth of field, denoising, and downsampling. These conditions should influence simulation and meshing decisions before the expensive cache is created.

Prepare stable sources and collision geometry

The render asset should not automatically become the simulation asset. Labels, threads, decorative bevels, nested packaging surfaces, and hidden polygons add collision cost without necessarily improving fluid behavior. Create dedicated proxies that preserve only the openings, wall thickness, contact surfaces, and edges capable of redirecting the liquid.

Build collision proxies that survive voxelization

Inspect the complete liquid path through the glass, bottle, cap, rim, base, tabletop, and any submerged objects. A model that looks closed in the viewport can still produce an invalid collision field because of open boundaries, inverted regions, self-intersections, or walls thinner than the collision resolution.

Utilities such as PolyDoctor, Clean, Fuse, and VDB From Polygons can support validation, but conversion should not be expected to repair fundamentally incorrect topology. Visualize the resulting signed distance field and inspect narrow necks, glass walls, sharp corners, and unintended internal cavities. If an important feature disappears at the effective voxel size, simplify or enlarge it in the simulation proxy while retaining the original geometry for rendering.

Ice deserves the same separation between render and collision assets. Faceted or highly detailed render geometry can be reduced to stable collision volumes, provided the planes and edges responsible for redirecting the drink remain intact. The material and lighting requirements are covered separately in this guide to photorealistic CGI ice.

Animated colliders require coherent velocity data. Avoid generating velocity in one transform space and moving the geometry again downstream. For fast motion, verify the collider at intermediate times and increase temporal sampling when necessary. More FLIP particles cannot repair a collider that tunnels between solver evaluations or supplies incorrect velocity.

Design the source around the shot

Use a dedicated source rather than emitting from a complex product mesh. Volume sourcing is convenient for continuous pours and broad initial masses. Particle sourcing offers more explicit control over birth position, separation, velocity, and per-particle attributes, making it useful for narrow streams, short bursts, or carefully designed initial states.

Place the source close enough to establish believable contact, but not inside the collision volume. An intersecting source can create immediate pressure spikes, trapped particles, or explosive spray that artists may mistakenly try to fix with damping. Visualize both the source volume and collision SDF together on the first simulation frame.

Source velocity should describe the dominant motion. Apply noise sparingly and at a scale appropriate to the liquid. Broad directional variation can help establish an asymmetric sheet; high-frequency random velocity across the whole source usually damages the primary mass before a readable crown has formed.

Keep the active domain close to the expected splash envelope while leaving enough room for approved droplets and retimed motion. An oversized domain increases pressure-grid and cache costs without adding surface detail. If the action moves substantially through the scene, consider a controlled moving domain only after confirming that it will not clip fast secondary elements.

Configure the Houdini FLIP simulation for readable motion

There is no universal hero-splash preset. Particle separation, substeps, viscosity, surface tension, and reseeding interact with source scale, collision speed, and the feature sizes required by the camera. Change one class of variables at a time and compare wedges over the same impact interval.

Choose resolution from the smallest required feature

Particle separation should be driven by the thinnest sheet, crown rim, or droplet that must survive—not by the overall size of the container. A finer separation can represent smaller structures, but it increases particle counts, pressure-solve cost, meshing cost, and storage. It can also reveal unstable high-frequency motion that was hidden in a coarse preview.

Begin with low-cost blocking that proves the source, impact direction, timing, and broad silhouette. Raise resolution only after those elements are approved. Preserve the blocking cache and settings so that changes introduced by the detailed solve can be distinguished from accidental changes to the setup.

Local or secondary solutions are often more efficient than globally reducing particle separation. If one camera-facing tendril or hero droplet needs special treatment, it may be better to protect or construct that element downstream than to multiply the cost of the entire simulation domain.

Balance substeps, reseeding, viscosity, and surface tension

• Substeps: increase temporal sampling when sources, colliders, or particles move too far between evaluations. This is particularly important for fast impacts, narrow streams, and thin collision regions. Substeps improve time integration; they do not add spatial detail.
• Reseeding: use it to maintain particle coverage where the liquid stretches or compresses, but inspect its effect on sparse sheets and isolated droplets. Aggressive reseeding can introduce changing point distributions that later appear as noisy or unstable surfaces.
• Viscosity: use it to resist deformation and preserve cohesion in thicker drinks. Excessive viscosity suppresses the crown, slows sheet extension, and can make a beverage feel heavy or gelatinous.
• Surface tension: use it to support rounded droplets and coherent rims. Too much can contract thin sheets into ropes or pull designed openings closed; too little can encourage diffuse breakup and irregular droplet shapes.

Evaluate viscosity and surface tension in the context of the product. Water-like drinks, juices, syrups, dairy beverages, and sauces should not share the same breakup pattern. Very viscous materials require a different solver and visual strategy; this Houdini chocolate fountain workflow illustrates the contrasting demands of a heavier fluid.

Do not use fluid parameters merely as emergency shape controls. If the crown is too low, first inspect impact velocity, source mass, collision direction, and timing. If the result explodes, check source–collider intersections and inconsistent velocity fields before adding viscosity or damping.

Art-direct the primary silhouette before adding spray

Impact energy comes from relative motion between the liquid and its collider. Review source and collision velocity together. A mismatch can produce a weak impact, an unexplained burst, or sliding at the contact point.

Shape the main event with source geometry, emission duration, and physically motivated velocity first. Add localized forces only where the physical setup cannot deliver the required composition. Masks based on position, time, or proximity to the impact are preferable to global turbulence because they preserve the large-scale direction of the splash.

The maximum-pose frame should remain legible as a still image. Check the crown height, sheet direction, negative space around the product, and the relationship between liquid and label. If these elements are weak, adding more droplets will increase complexity without fixing the shot.

Separate secondary droplets from the primary liquid

Once the connected body works, classify or derive secondary elements using velocity, acceleration, age, position, connectivity, distance from the main body, or custom impact masks. Secondary particles should inherit meaningful fluid velocity rather than receiving uniform random motion. This keeps their trajectories connected to the event.

Limit emission spatially and temporally. A short window around the impact usually reads more clearly than continuous spray. Delete or fade particles after they leave the useful composition, and omit details hidden behind the product or reduced to subpixel highlights.

It is often useful to maintain three controllable classes:

• Primary mass: connected liquid that carries the silhouette and product contact.
• Breakup pieces: medium forms that bridge the crown, sheets, and isolated droplets.
• Fine spray: sparse accents used for energy, depth, and edge detail.

Keep these classes available as separate cache or render branches. Their density and shading can then be revised without destabilizing the approved primary simulation.

Cache and retime without losing continuity

Cache FLIP particles before meshing. The particle cache should include stable identifiers and the attributes required downstream, commonly velocity, age, life, scale, and custom classification masks. Remove unnecessary attributes only after confirming that they are not required for meshing, secondary generation, motion blur, or handoff.

Retiming is usually safer at the particle stage than after polygon meshing, provided the particles can be interpolated consistently. A topology-changing mesh is difficult to interpolate without popping. After retiming, recalculate or validate velocity for the new time scale rather than assuming the original vectors remain suitable for motion blur.

Moderate retiming can improve the presentation of an approved event, but it does not preserve every visual relationship automatically. Slowing the action changes apparent droplet spacing, breakup duration, gravity cues, and blur length. Large timing changes may require a new simulation because the revised action no longer looks like the same physical event.

Review the retimed particles, final mesh, and render at the delivery frame rate. Missing source frames, inconsistent IDs, cache boundaries, or incorrect velocity scaling may remain invisible until motion blur is enabled.

Mesh FLIP particles while preserving thin features

Meshing is a distinct creative and technical stage. A strong particle simulation can be damaged by excessive smoothing, while a noisy particle field cannot always be rescued by a finer surface. Cache the particles first and branch the meshing network so that reconstruction settings can be revised independently.

Choose a reconstruction method

Particle Fluid Surface is a practical starting point when particle coverage is reasonably uniform and an integrated reconstruction workflow is sufficient. It provides direct controls for particle scale, filtering, smoothing, and polygon output.

A VDB workflow—typically rasterizing with VDB From Particles, processing the field, and converting it to polygons—offers more explicit control over the intermediate volume. It is useful when the surface requires targeted dilation, erosion, masking, or region-specific filtering. The trade-off is that broad field operations can close narrow gaps, fuse droplets, thicken sheets, and remove delicate rims.

Neither method can recreate a stable feature that the particle cache never represented. Before increasing mesh resolution, inspect the points in the shot camera. If a sheet has no coherent particle support, repair the simulation or build a controlled secondary element instead of expecting the mesher to invent it.

Use separate treatments for different feature classes

The primary body and isolated droplets rarely need identical filtering. A useful workflow is to classify them before reconstruction, mesh the main liquid with enough filtering to remove particle-scale noise, and process selected droplets with a smaller influence radius or dedicated geometry branch.

Apply the least smoothing that creates a continuous, renderable surface. Judge it at final or near-final output size. A surface that appears slightly rough in a close viewport may render convincingly, while an immaculate viewport mesh can lose every crown edge and thickness transition after broad smoothing.

Maintain at least three comparable outputs:

• a minimally filtered diagnostic mesh;
• the production mesh;
• any aggressively cleaned or art-directed alternative.

This makes it possible to identify whether a missing detail originated in the simulation, particle classification, field reconstruction, polygon conversion, or smoothing stage.

Check temporal behavior, not only hero frames

Inspect the entire frame range for pinching, holes, bridging, faceting, and topology flicker. Thin sheets are especially vulnerable because small changes in particle distribution can open or close the reconstructed field. Nearby streams may merge if particle influence or smoothing is too broad, while isolated droplets can disappear when minimum-size filters are too aggressive.

Stable geometry can still produce crawling highlights if normals, refraction, or roughness respond sharply to subtle topology changes. Compare the particles, raw mesh, filtered mesh, and neutral-material render before returning to the solver. This isolates simulation instability from surface or shading instability.

Build a beverage shader that reads through thickness

A beverage should not depend on a saturated surface tint alone. Its identity comes from the combination of reflection, refraction, absorption or volume attenuation, roughness, and geometry thickness. One material may cover the complete splash, but thin sheets, thick masses, and droplets can require controlled variations within the same optical family.

Use physically plausible refraction as a foundation, then tune attenuation so thicker regions develop appropriate color depth while thin sheets retain transmission. If attenuation is too strong, the body becomes opaque and the edges dirty. If it is too weak, the drink reads as clear glass regardless of its intended color.

Thin sheets often need deliberate edge contrast from lighting, roughness, or restrained attribute-driven variation. Avoid solving visibility solely by increasing saturation or reflection strength; those changes may make the splash look metallic or plastic. Test the material against both the intended background and a neutral diagnostic background.

Bubbles, carbonation, foam, suspended pulp, and condensation should be separate, controllable systems when they materially affect the image. Their size and density must make sense at the final camera distance. Detail that resolves only as noisy pixels should be reduced rather than treated as proof of realism.

Light and render the splash with the product

Transparent liquid is defined by what it reflects and refracts. Large side or back sources can create continuous highlight bands across crowns and droplets, making the silhouette readable without filling the set with small, competing reflections. The best setup is the one that separates the liquid while preserving the packaging, logo, and product form.

Evaluate overlaps carefully. A bright liquid rim crossing a label may improve the splash in isolation while damaging the advertisement. Adjust light size, direction, camera angle, and background value before forcing separation through extreme material settings.

Validate the actual scene in Karma or the selected production renderer. Overlapping glass, beverage, splash, ice, and background surfaces can create noise, excessive internal darkness, confusing refraction layers, or fireflies that are not visible in the viewport. Render representative frames containing the thickest mass, thinnest sheet, highest droplet density, and most complex product overlap.

Treat motion blur as part of shot design

Motion blur should be tested while the simulation is still adjustable. A fast droplet may become a useful directional streak, disappear across too many pixels, or merge with another highlight. Too little blur makes high-speed particles feel disconnected and frozen; too much can erase the crown rim and impact point.

Use the production shutter, frame rate, velocity configuration, depth of field, sampling, and denoising settings during validation. A feature is not approved merely because it exists in the cache—it must survive the final imaging pipeline.

Render diagnostic outputs for the liquid body, droplets, foam or bubbles, container, reflections, refraction, volume contribution, motion vectors, depth, and object mattes where the renderer supports them. These passes reveal whether missing detail is caused by geometry, transparency, lighting contrast, motion blur, or compositing.

Diagnose common beverage splash failures

Increase resolution only after identifying the failing stage. Higher particle counts will not repair an intersecting source, invalid collision field, unsuitable mesh filter, or unreadable lighting setup.

Failure	Likely causes	What to inspect first
Collision leaks	Thin or open proxy, insufficient temporal sampling, source inside collider, invalid collider velocity	Visualize the collision SDF, source overlap, intermediate collider positions, and transform space
Explosive breakup	Excessive source velocity, source–collider intersection, broad random velocity, inconsistent animated collision velocity	Disable secondary forces and inspect the first contact frames
Dead or heavy motion	Excessive viscosity, damping, weak impact energy, coarse source shape	Compare source and collider velocities, then test fluid properties independently
Disappearing sheets	Feature below particle resolution, sparse coverage, aggressive reseeding behavior, excessive mesh filtering	Compare particles, raw surface, and filtered surface through the shot camera
Blobby mesh	Large particle influence, coarse field, excessive smoothing, fused feature classes	Run a short mesh wedge and process body and droplets separately
Fragmented or noisy mesh	Sparse particles, insufficient filtering, unstable spray included in the hero surface	Inspect particle support and classify secondary elements before meshing
Crawling highlights	Topology changes, unstable normals, sharp roughness response, high-frequency surface noise	Render raw and filtered meshes with a neutral material, then restore the beverage shader
Liquid resembles colored glass or plastic	Flat surface tint, weak attenuation, unsuitable reflection balance, poor lighting contrast	Test thickness-dependent absorption, refraction, roughness, and large-source reflections
Droplets vanish in the final render	Subpixel scale, excessive blur, depth of field, denoising, low contrast	Validate at delivery resolution with the production camera and shutter
Unmanageable caches	Oversized domain, unnecessary attributes, globally excessive resolution, uncontrolled secondary particles	Constrain the domain, separate hero and secondary detail, and review cache contents

A structured diagnostic sequence prevents unrelated changes from masking the cause:

1. Review source and collision fields at the failing frames.
2. Inspect the particle cache without meshing.
3. Render the raw mesh with a neutral material.
4. Compare primary liquid and secondary droplets separately.
5. Restore the beverage shader under controlled lighting.
6. Enable motion blur, depth of field, denoising, and compositing at delivery resolution.

Version and package the production shot

Simulation, meshing, secondary effects, look development, and lighting should be versioned independently. A shading adjustment should not overwrite an approved FLIP cache, and a new mesh filter should not make the originating particle cache ambiguous.

A clear dependency chain might distinguish:

• source_v012 for emitter and collision geometry;
• flip_sim_v006 for particles generated from that source;
• mesh_v004 for the primary surface and droplet geometry;
• secondary_v003 for spray, bubbles, foam, or mist;
• lookdev_v009 and lighting_v005 for render-stage revisions.

The exact convention can vary, but every output should identify its upstream dependencies, frame range, take, and version. Preserve the Houdini scene, cache paths, renderer and plugin versions, required HDAs, color-management configuration, and project variables. Paths must resolve in a clean render environment rather than only on the originating workstation.

Record the approved technical state

Shot metadata should include scene units, frame range, time scale, particle separation, substeps, reseeding state, viscosity and surface-tension controls, meshing resolution, retiming method, render resolution, shutter, and camera-specific overrides. Also document deliberate cheats such as enlarged collision walls, protected droplets, localized forces, or camera-facing geometry.

Keep wedge results traceable. Name the chosen test rather than silently replacing it, and preserve enough preview information to compare changes in impact timing, crown shape, breakup, cache size, and render behavior. This is particularly valuable when a more energetic variation creates excessive spray or loses the approved product silhouette.

Validate the handoff

Before delivery, scrub the full particle and mesh ranges for missing frames, abrupt count changes, invalid geometry, collision leaks, unstable thin features, and cache-boundary discontinuities. Then render representative final frames at the actual output format with production motion blur, color management, and compositing passes.

Compositing can refine contrast, bloom, mist, restrained spray, shadows, and background separation. It should not be expected to repair a missing liquid contact, a leaking collider, or a crown that needs correct geometry to refract its environment.

The final package should include a concise dependency map showing which source generated the FLIP cache, which cache generated each mesh, which secondary systems depend on those outputs, and which camera and render settings reproduce the approved image. With those relationships intact, another artist can revise timing, droplet density, meshing, or look development without sacrificing the approved hero splash.