Have you ever felt overwhelmed by the sheer complexity of achieving lifelike CGI water effects?
When working with Houdini, you know that fluid sims can be a technical maze. High-resolution splashes, mesh cleanup and caching all demand precision and patience.
Does the challenge of replicating natural water behavior slow down your pipeline? Small settings tweaks can cause ripples across your entire scene.
Designing for a brand like Biotherm raises the bar even higher. Their emphasis on purity and performance means every droplet must look and feel real.
Here we bridge brand expectations and technical execution to help you set up robust sims without sacrificing visual appeal or control.
We’ll explore the role of science in fluid dynamics, share key parameters for realistic motion, and highlight ways to optimize render times.
You’ll discover practical setups, caching strategies, and lighting tips so your next motion design water sequence runs smoother and renders faster.
What were Biotherm’s scientific and creative objectives for the CGI water campaign?
Biotherm tasked our team to merge rigorous scientific accuracy with an emotive visual language. On the scientific side, they needed molecular-scale fidelity in fluid dynamics to reflect plankton-infused extracts. Creatively, the goal was to evoke purity and motion, turning each droplet into a dynamic narrative that emphasizes product efficacy.
The primary scientific targets included:
- Reproducing micro-surface tension effects via Houdini’s FLIP solver and particle resize to capture droplet pinch-off and coalescence.
- Simulating thermal gradients with custom VEX fields that map temperature variations to subtle color shifts in the fluid.
- Bridging macro and micro scales by layering high-res openVDB meshes over low-res guides, preserving detail without exploding sim times.
On the creative front, Biotherm aimed to:
- Create slow-motion reveals of encapsulated product ingredients using RBD-driven collisions to shatter water skins around active cores.
- Integrate mist and vapor accents through a sparse Pyro solver, giving scenes an ethereal, science-driven atmosphere.
- Maintain a clean, brand-aligned aesthetic by leveraging procedural shader networks for consistent refractive and specular highlights.
To align both objectives, we developed a modular Houdini Digital Asset that exposed key parameters—viscosity, surface tension, color ramp—allowing rapid iteration while ensuring every frame met Biotherm’s standards for scientific authenticity and visual impact.
How was Houdini embedded into the production pipeline for water, science visuals, and motion design?
Integrating Houdini into Biotherm’s pipeline required a modular approach that aligned with existing asset management and scheduling systems. We established three parallel tracks—water simulation, scientific volume rendering, and procedural motion design—each feeding into a unified USD stage. This ensured consistency and facilitated iterative updates without disrupting downstream compositing or editorial.
At the core, production asset files were published as USD pods via Python-driven LOPs. Each pod contained geometry references and metadata tags for resolution, simulation seeds, and material assignments. This method allowed lighting artists to override shader parameters in Solaris while preserving the upstream simulation cache structure.
- Source geometry and emitter setup defined in SOPs, versioned in Perforce
- FLIP Solver networks parameterized with shelf tools extended by HDA wrappers
- PDG (TOPs) used to dispatch hundreds of sim variants across the render farm via HQueue
- Whitewater, mist, and droplet passes cached into .bgeo.sc for selective playback
For scientific visuals, we imported CSV and volumetric data sets through SOP Import nodes, converted to VDBs, and ran custom pyro advection to visualize fluid flow and diffusion patterns. This process leveraged POP Advect fields within a DOP network, with each timestep written out as multi-channel VDBs. Artists could swap data sources at any time without rebuilding the network, thanks to parameterized filepath expressions and wildcards.
Procedural motion design elements—such as animated waveforms, abstract fluid ribbons, and particle trails—were authored using CHOPs for temporal modulation and KineFX for geometry retargeting. Each procedural rig was encapsulated in an HDA, exposing only high-level controls (speed, frequency, amplitude) to maintain consistency across shots. These rigs streamed into the Solaris stage as instanced geometry, ensuring memory efficiency.
A master PDG workflow orchestrated the end-to-end process: fetching input asset versions, running sims, generating preview caches, updating USD stages, and triggering Karma XPU renders. This approach delivered automated error checking, notification emails on failures, and snapshot reports of simulation statistics. By embedding Houdini at multiple pipeline layers—from simulation to final render—we achieved a robust, scalable infrastructure that supported last-minute changes while preserving scientific accuracy and visual fidelity.
Which Houdini solvers, setup patterns and multi-scale strategies produced Biotherm’s photoreal water?
Multi-scale FLIP + particle/spray coupling: seeding, resimulation and high-res mesh generation
We built a low-resolution base using Houdini’s FLIP Solver inside a DOP network, tuning particle separation to capture large waveforms. A secondary POP network seeded spray, foam and droplets based on velocity divergence and surface curvature. That coarse sim provided guide velocities, rest positions and color attributes for a high-res resimulation.
For high-resolution detail we imported guide fields into a fresh FLIP DOP, dropped particle separation by 50–75%, and used attribute transfer from low-res particles to seed new ones. The result is consistent small-scale splashes without re-computing large motion.
Finally we generated a dense mesh by converting particles to VDBs, remeshing and subdividing in SOPs. A point-based curvature smoothing SOP reduced noise, while adaptive UV scattering preserved filmic specular highlights on thin edges.
Surface tension and thin-film treatments: curvature forces, viscosity handling and secondary detail workflows
Accurate capillary effects relied on the FLIP Solver’s surface tension controls. We increased the curvature force weight and applied a bilateral filter on particle normals to avoid high-frequency jitter. A custom VEX force node further boosted tension near contact points, emulating water beads on product surfaces.
- Viscosity: per-particle pseudo-viscosity via the viscosity tab, blended with a small Gaussian filter in SOPs
- Thin-film: generated a secondary shell by offsetting the surface VDB and converting it to a thin mesh layer
- Refraction wrinkles: used a curvature-driven noise SOP to modulate normals on the thin film
Secondary workflows for bubbles and fine mist utilized Houdini’s Whitewater shelf tools inside a POP network. Emission rates were driven by collision impact and local curvature thresholds, then fed into a particle-to-mesh workflow for micrometer-scale highlights.
How was look-development and physically based shading executed to convey water, skin science, and Biotherm branding in renders?
In Houdini’s look-development stage, each material was built inside a /mat subnet to isolate fluid, dermal, and product shaders. For physically based shading of water, we used a Principled Shader with IOR locked at 1.333, dual-lobe GGX specular, and a thin-film interference node to capture micro-droplet sheen. A VEX-driven normal blend integrated our FLIP sim’s surface turbulence into the final shading network.
To simulate skin science, we layered a low-scatter diffuse pass over a multi-scatter SSS setup using Houdini’s Skin Microflake node. Diffuse albedo, specular roughness, SSS radius, and porosity were packed into UDIM channels. Transmission and subsurface weights were driven by photogrammetric color data, preserving pore detail, then fine-tuned via a custom random-walk SSS VEX snippet to match measured biophysical profiles.
For Biotherm branding, a subtle rim-light gradient tinted with the brand’s blue Pantone was composited into the shader’s emission input. A floating water-film layer reused the base IOR but applied animated UV noise derived from FLIP velocity fields—this linked fluid motion to a cosmetic gloss effect. Final renders were color-managed in ACEScg to ensure consistency across all beauty and product shots.
How were scientific datasets and lab visuals integrated, validated and communicated to stakeholders (data-driven sims, colorimetry, acceptance criteria)?
In our Biotherm pipeline, the first step involved ingesting measurement tables and volumetric scans from lab experiments. Using Houdini’s File SOP and Attribute Wrangle, raw CSVs and PLY point clouds were converted into attribute volumes. We preserved temperature, salinity and particle concentration channels as VDB fields, laying the groundwork for truly data-driven simulations.
Once the volumetric fields were loaded, we parameterized our FLIP solver via Python expressions tied to lab metadata. Velocity magnitudes and density gradients drove viscosity and surface tension attributes. By referencing experimental Reynolds numbers directly within the solver’s physical properties, each simulation run mirrored real-world fluid behavior without manual curve fitting.
Color fidelity was achieved through a dedicated colorimetry pipeline in Houdini’s COPs. We converted lab-captured swatches into CIE LAB space, then mapped them into our render’s linear RGB workflow via a custom LUT generated with OpenColorIO. By embedding gain and gamma adjustments at the shader level, renders matched microscope images within a ΔE00 tolerance of 1.5.
To validate results against lab visuals, we implemented automated histogram matching and spatial correlation metrics. Volume overlaps were quantified by comparing VDB density centroids, while surface silhouettes used edge-extracted curves. We defined acceptance criteria that ensured deviations stayed within engineering tolerances for Biotherm’s R&D team.
- ΔE00 color difference < 1.5
- Volume centroid displacement < 2 mm
- Shape similarity index ≥ 0.90
- Velocity profile RMS error < 5%
Stakeholder communication leveraged side-by-side flipbooks and interactive flip panel web viewers. Each deliverable included embedded charts showing metric convergence, annotated render passes, and inline comparisons against microscope stills. This transparent, metric-driven presentation fostered rapid acceptance from both scientists and creative directors.
What pipeline optimizations, automation and asset management practices guaranteed performant sims, scalable rendering and on-time delivery?
To support a high-volume campaign like Biotherm’s water simulations, we structured a modular pipeline in Houdini that prioritizes cache locality, task parallelism and robust version control. This approach ensured both interactive iteration and reliable farm throughput.
- Procedural Caching Architecture: We encapsulated each sim stage—source emission, collision preprocessing, post-turbulence—in separate DOP networks. Using a standardized ROP Fetch TOP network, caches write to a structured directory: /cache/$SHOT/$STAGE/$VERSION.bgeo.sc. This isolation lets artists update one stage without invalidating upstream data.
- Dynamic Memory Management: Within FLIP simulations we enabled “Auto-Resize Fluid” and custom shrink-wrap bounding boxes, capping grid volume via frame-based expressions. This cut voxel counts by 30% while retaining splash fidelity.
- Parallel Task Orchestration: Leveraging PDG’s dispatch to dispatch FLIP sim slices and meshing tasks across cores, we aggregated results with a late-merge TOP chain. Each frame’s meshing ROP ran simultaneously, reducing wall-time by 60% compared to linear dispatch.
- Automated Quality Gates: Python callback scripts on ROP triggers perform checksum validation and check grid artifacts against a reference clip. Failing frames auto-flag in the tracking system for immediate review.
- Digital Asset Versioning: All key setups—emitter rigs, solver setups, render LOPs—were wrapped as HDAs with locked non-essential parameters. We tracked versions in Perforce, with auto-incremented changelists on export. This enforced reproducibility and rollback capability.
- USD-Based Render Prep: Output geometry and volumes were converted to USD at sim completion. LOP networks then instanced Biotherm product models and scattering setups, enabling look-dev and lighting teams to iterate without reloading heavy SOP networks.
By combining structured caching, PDG-driven concurrency, automated QA checks and strict asset version control, we achieved a pipeline that scales seamlessly from local workstations to render farm, delivering high-resolution water sims on schedule with minimal manual intervention.
