Introduction
Context
Advancing technologies within the VFX industry have brought us rapidly closer to recreating photorealistic animals and creatures. Digital Grooming is the part of that process that focuses on creating dynamic simulations to replicate fur and hair for digital creatures. Breakthroughs and advancements in rendering technologies and lighting techniques, like ray-tracing and dynamic interpolation of guide hairs, have allowed artists to continuously bridge the gap closer to photorealism.
Research Focus
The research will focus primarily on how grooming has advanced within the VFX industry and the role audience perception plays in the production workflow. This will tie into the actual workflow of procedurally generating a groom in Houdini in order to better understand the processes that are implemented. Since the industry is always evolving, new techniques and advancements will also be covered to identify the direction the industry is taking.
Rationale
Understanding the process required for digital grooming and the physics behind rendering photorealistic hair is fundamental to generating a believable groom and improving audience perception. New tools and software applications are being constantly developed, with rising demand for real-time workflows driving the way and pushing the industry forward. To have a successful career as a grooming artist, one needs to stay constantly up to date and constantly pursue new and relevant workflows to diversify their skill set.
Aims and Objectives:
The main aim of this investigation will be to evaluate how different environmental conditions can affect the condition of a creature's groom. This will also be analysed against the perceived level of realism across the various levels of environmental effects and how this compares to a clean groom.
To achieve this, one must gain an understanding of the workflow and pipeline typically involved in the grooming process and how large-scale studios are driving the way forward and offering artists fine control whilst working in a procedural workflow. The practical experiments will utilise the researched techniques and theories to create realistic grooms to industry standards and present them to a panel of professionals as well as audience members to establish how realistically they are perceived.
Early Uses of Groom and the Methods that Developed
Digital grooming is a rather specialised job that is essential for any production featuring digital characters or creatures. Groom artists are tasked with recreating fur and hair for use on digital characters in line with the director’s vision, whether that is stylised or photoreal. Their responsibilities lie both in the creative art direction of the groom as well as the technical processes like optimisation and simulation interactions. This includes tasks, such as the creation of fur, hair and feathers for characters and creatures, collaborating with Creature Technical Directors to achieve solutions for required dynamics, as well as collaborating with departments like texturing and Look-Dev to achieve the desired aesthetic (Industrial Light and Magic 2025). Given the amount of technical knowledge and experience that is required for this type of role it can be very well paid, though this depends on location and studio, with some artists at ILM in Vancouver being offered a potential salary of $159,000 CAD (£86,100) and up to £82,000 in the UK, though this varies vastly across different jobs (Industrial Light and Magic 2025; Glassdoor, 2025).
There are a few different technical approaches when it comes to digital grooming; each one has different advantages and disadvantages based on the context of the production. The main methods applied in modern-day productions are Guide-based Grooming, Procedural generation and Hybrid workflows that combine the two.
Figure 1: Early use of digital fur grooming, pioneered by ILM for Jumanji (1995)
Guide Grooming is the oldest and most thorough method employed, having its origins back on the set of Jumanji (1995), where ILM were pioneering the way forward with their CG Lions and Monkeys. To be able to achieve the creative aspirations set out by the directors, the team at ILM came up with a proprietary software which allowed them to replicate the Lion animatronics’ fur digitally, by using a series of control hairs to art direct it before using linear interpolation to cover the rest of the lion (Forrest, et al. 2022). As demonstrated in Jumanji (1995), guide grooming generally involves creating a series of control hairs across the model, which are evenly spaced out, to act as a guide for the grooming and styling process. A groom artist will then use various brushes and tools to manipulate these hairs and style them according to the desired effect before using strand interpolation between the guide hairs to cover the remaining model (NVIDIA 2016). This method of purely working with guide grooming is a very manual and time-consuming process which also ultimately ends up affecting the photorealism of it. A purely manual based workflow and have the effect not introducing enough natural fur breakup and clumping nor adding simulated effects. Creating a wet or windy groom this way has diminished returns as an artist can only so accurately replicate physical simulations. As seen in Figure 1, this can result in flat looking fur with and diminished perception of realism for general audiences.
Figure 2: Grooming Density map used for hair interpolation by NVIDIA.
To gain more nuanced aesthetics, texture maps were created to drive various parameters based on their UV coordinates. Figure 2 shows an example of a grayscale density map, which controls how many hairs are generated in each area; the lighter the area, the more hairs are created. These kinds of maps were used because they allowed for greater control and flexibility when creating a groom, as well as giving more natural styles through the use of noise patterns, which would be painstaking to replicate manually (Baker & King 2003). The advantages of using guide grooming are that it allows artists to have prominent levels of creative control over the groom as they can manually determine which way the hair should flow, which can be particularly necessary on stylised productions. It also offers vast performance improvements over other methods since the software does not have to try to simulate hundreds of thousands of hair strands, dynamically interacting with each other, which would typically take vast amounts of computing power and RAM storage to process (Traynar, et al. 2025). However, this can result in a groom appearing ‘too’ stylised since it's harder to create naturally appearing imperfections like clumping and frizz, as well as being incredibly time-consuming.
Figure 3: Full body groom on a hero asset by Weta FX for King Kong (2005)
As CGI was being implemented more frequently and computer processors advanced, artists began to create new techniques to tackle digital fur. This caused the advent of procedural grooming as an alternative to linear interpolation. One of the pioneering productions for this was Peter Jackson’s King Kong (2005), which tried to solve the issue of how to accurately represent and render the millions of hairs on the giant King Kong model as seen in Figure 3. To be able to accomplish this, Weta FX had to create bespoke software, now dubbed ‘BarberShop’, to be able to procedurally generate up to 30,000 – 40,000 different clumps of hair which can dynamically interact with an environment, adding imperfections like leaves and mud as Kong moved around (Failes 2017). This kind of procedural grooming is a non-destructive workflow, using layers of various nodes with different parameters to generate dynamic fur. Rather than manually painting out guide geometry and interpolating it, the software uses point scattering to generate geometry across the mesh with adjustable parameters based on curves and noise patterns, which control attributes such as clumping and imperfections (Capper 2016). This dynamic workflow is much more efficient in terms of manual labour required and the number of creative decisions that can be made. If a creative decision gets altered further upstream whilst the groom is already in process, it is much easier to revert and implement changes into the groom compared to manual guidance. Though this workflow does not come without its drawbacks, as it takes tremendous amounts of processing power to be able to render all that fine detail, adding to electrical and hardware costs. This is why most modern VFX studios tend to use hybrid workflows that make use of guide geometry to artistically drive the groom prior to using procedural generation to fill in the fine hairs and add natural nuance to the groom.
Technological Breakdown of Modern Solutions and Limitations
For modern productions, most major studios like Weta FX, Sony Image Works and ILM have developed proprietary solutions for grooming that are more tailored to their specific needs. One such example would be Weta FX’s ‘BarberShop’ and ‘Wig’ applications, which are proprietary packages for creating fur and hair, with ‘Wig’ specialising in creating bespoke characters wigs and headpieces. ‘Wig’ is an example of a destructive workflow since the hairs are not simulated, giving artists full control over each one (Weta FX 2015). The two main standardised and publicly available software packages for grooming would be Maya’s XGen and Houdini, with its vast suite of procedural grooming tools.
Figure 4: User interface for Maya's XGen interactive groom.
Maya’s legacy XGen is an example of a destructive guide-based workflow for grooming. Rather than generating procedural and dynamic hairs, XGen will instance the guide geometry to fill out the remainder of the groom, creating physical geometry for each hair (Autodesk 2015). It also offers a dynamic range of brush-based tools for styling hair and fur, giving artists full control over the artistic aesthetic. As seen in Figure 4, XGen has also developed a procedural suite of tools, offering interactive grooming workflows via a node-based system that is rendered in real time on the GPU, offering a non-destructive alternate workflow (Autodesk 2017). This is often used for more stylised or smaller-scale fur projects as it does not have dynamic interactions and only a limited range of procedural tools, still mainly relying on artistically driven brushes.
Figure 5: Example workflow for grooming a Cat in Houdini
Its counterpart, Houdini, has a much more fleshed-out and robust ecosystem of procedural tools that allow for powerful manipulations in a completely non-destructive workflow. It blends between using artistic brushes with non-destructive procedural generation, meaning that if artistic changes are made upstream, the entire groom gets updated without destroying the generated mesh and manual work (Coetzer 2022). As seen in Figure 4, this is a result of its node-based system allowing for layering of effects and various parameters, giving it such flexibility.
Figure 6: Using the Guide Process SOP and Hair Generate SOP for creating hair.
Houdini’s workflow starts off similar to XGen by generating guide geometry across a mesh. This is typically done in localised areas using attribute painting since different areas of a creature have varying hair types and attributes. For example, hair found in the ears of most animals is generally much shorter and finer than the rest of the body, which would have a thicker coat. Artists can then use brushes to manipulate the flow and direction of the guide hair, using reference materials to get an accurate representation. Houdini also offers procedural control over parameters, such as guided deform, clump, length, frizz and masking, via the Guide Process SOP node (shown in Figure 6), which can be used to manipulate the style and flow of guide geometry by adjusting varying attributes as well as using noise and texture maps to drive randomised patterns (SideFX 2025). These kinds of operations are what drive up the level of perceived realism in a digital groom as they offer artists ways to dynamically generate and replicate natural occurring effects. Audience perception of realism depends on their preconceived notion of what that animal should look like, within their natural habitat. You would rarely find a perfectly clean groom of a monkey or a wolf in the wild. Animals are messy, they have dirt stuck between their hairs, damp fur sticking together and clump at random angles. Procedural operators in Houdini allow artists to emulate these environmental effects whilst still maintaining artistic control, to improve photorealism.
Figure 7: Inside the Guide Deform SOP
Grooming is always done on a static mesh, though the Guide Deform SOP in Houdini can be used to animate the guide hairs. To achieve this, one needs to plug in the guide geometry into the first input, as well as the static mesh underneath and then the animated mesh, to inform the node how it should be deforming the hairs. This is essential when working with an animated creature or character since it attaches the guide hairs to the deforming mesh, allowing it to follow body movements without needing complex simulations (Side FX 2017). For dynamic interactions and simulations, Vellum Hair can be utilised, allowing for physics-based reactions to forces such as wind and water via the use of a POP advect by Volume node, which allows the Vellum objects to be influenced by the velocity of a FLIP fluid simulation without direct interaction (Side FX 2025). The hair generated from a Hair Generate SOP is ready to be textured, shaded and rendered using Karma, Solaris, or any other render solution compatible with Houdini.
Figure 8: Texturing and Shading of a groom for a cat in Houdini.
The texturing process for a groom in Houdini is a fairly standard workflow when compared to other pipelines, though the main differing factor is the focus on using attributes as well as texture maps. The colour is driven by the Cd attribute or a pre-made texture map to introduce variation and achieve a more realistic look. Other attributes like width, length, clump, and U position all inform the shader settings used for texturing certain areas that might require length-specific opacity shading or adding colour variations for clumping. Hair is anisotropic, which means that it reflects light along the strand direction rather than evenly in all directions, with the light orientation and intensity as well as the camera angle severely affecting the perceived output (Marmoset 2025). Houdini achieves this by rendering hairs as curves, not geometry, preserving strand tapering as well as supporting anisotropic shaders to accurately represent this phenomenon.
Whilst Houdini remains as one of the most versatile and dynamic solutions for creating a photorealistic groom, it is not without its Limitations. One of the mains being how computationally intensive it is, particularly at high-density strand counts, requiring substantial amounts of RAM, so simply viewing the simulation in the viewport. This is due to the fact that hair representation is rather complex due to the way it interacts with light, resulting in complex simulation systems which are still challenging for modern systems to process (Luo, et al. 2024). This can make having a considerable number of groomed hero assets very costly, requiring large render farms to render out the simulations efficiently, which is not always affordable or feasible for smaller studios. Therefore, large grooms like these are typically reserved for higher-budget productions where a producer can afford to splurge on a full procedural groom. Additionally, as previously mentioned, procedural grooming can lack fine control, making it hard to be artistically driven. This also stems from long cook times when making changes to a vast array of strands, which can significantly increase the time for feedback on creative decisions, slowing down the overall process (Andrus, et al. 2022). Large studios have developed proprietary solutions to combat this, like Weta FX’s ‘Wig’ and MPC’s ‘Furtility’, which work over the top of Houdini’s existing toolset. As a result of these limitations, the industry has begun researching real-time solutions, especially for game engines like Unreal Engine, due to the rise in demand for Virtual Production-compliant solutions. These methods would also greatly improve the photorealism of digital grooms for real-time applications like video games and virtual production. An improvement in procedural tools for real-time applications means that artists are able to more effectively simulate environmental changes within the digital landscape. This could include a storm in a cold, snowy Unreal Environment, with particles and wind interacting with the groom in real-time. This would sell the believability to the audience more, since it replicates the natural effects they would expect to see in that scenario. Creating a pre-baked groom would have limited opportunities for real-time interactions, reducing the integrity of the overall integration into the environment.
Advancements and Future Trends
The VFX industry is an ever-changing landscape with new advancements being made across all sectors, including digital grooming. The rise in demand for Virtual Production and real-time FX solutions has been a large driving force for the direction the industry is taking. Unreal Engine has been an industry leader in all things real-time, which now includes grooming workflows thanks to a new Hair-Raising tool, which was implemented in Q1 of 2025 (Beier 2025). This particular plug-in has features supporting interactive grooming, real-time adjustments, as well as importing external assets to adjust third-party designs. Its use case is meant primarily for Cinematic workflows and in-camera VFX applications like use on a virtual production stage. It also received a 2.0 update recently, further improving the shading capabilities by allowing for a more simplistic control interface, giving artists more flexibility and artistic direction (Thacker 2025).
Unreal also possesses native capabilities via the Hair Strand grooming system with a heavy focus on performance. It includes support for physical rendering properties like anisotropic attributes, translucency, and detailed shadows, as well as being fully integrated with their Meta-Human Framework (Morand, et al. 2021). These kinds of implementations allow for rapid look development of digital characters and creatures without the need for offline renders and long processing times. Though Unreal’s native solution does have some limitations, as it does not offer full control over the style of the groom once it's imported, which is where 3rd party solutions, such as Ornatrix, are so popular. Ornatrix is a well-established plug-in previously used primarily in Maya, which now offers procedural grooming within Unreal Engine. It offers fine control over grooming functions through various brushes and combs, as well as supporting procedural generation of frizz, curls, clumping and braiding (Ephere 2022).
With the implosion of generative AI, the question has also been asked how this might affect CFX artists and digital grooms. A research study, Li, Lin, & Zhu 2025, identified methods for physically based video diffusion models for dynamic hair rendering. As part of the methodology, they combine physics-based simulations and video diffusion models in an attempt to get the best of both worlds and improve the underlying issue with hair and generative AI. Hiar consists of thousands of individual strands, which all interact dynamically, making it rather complex to accurately represent. The overall conclusion of the investigation was that visual fidelity and detail improved, particularly during fast head movements and under external forces such as wind. However, whilst this seems to demonstrate a potential new workflow for generating human hair, it offers vastly diminished manual control and thus cannot compete with traditional grooming workflows. Hugh Johnson, a Senior Houdini artist at Framestore, revealed in an interview that he believes AI will never be able to match industry requirements for manual control, especially for high-end productions where visual detail is key. Generative AI also has large running costs associated with it, with some Gen AI models costing around £1000 per prompt (Johnson 2026). At a large scale, this means that current VFX workflows are cheaper and more efficient and also give artists and directors increased manual control. The majority of the details and the last 10% of the perceived level of photorealism come from manual adjustments, which take about 90% of the time, according to Johnson (2026). This shows that whilst AI models have a place for lower-quality work where visual fidelity is not as much of a concern, high-end productions will remain with traditional VFX workflows where audience perception is a key factor.
Conclusion
Grooming as a workflow in the VXF industry is continually adapting as technology advances. Artists require continuous training in a variety of pipelines to keep up with current industry standards. As the industry evolves, so does the audience's level of expectations for visual fidelity. Weta FX’s work on the Avatar films, among other productions, set the precedent for environmental interactions with digital grooms, especially water and fluid interactions on a larger scale. As photorealism reaches a visual crux, the emphasis now shifts to focusing on improving computational improvements, bridging the gap between artists' direction and audience perceptions of naturally occurring grooms. Successful grooms hinge on an artist’s ability to use all the tools at their disposal to accurately replicate the real-world reference, including studying the proposed environment and the naturally occurring environmental effects that would be entailed. AI and procedural methods have their place in the modern pipeline, but ultimately, manual control is still king when it comes to maintaining visual fidelity and adding environmental effects for the audience's sake.
Practical One
This first practical aimed at creating a clean groom using guide hairs in Houdini to establish a foundation for further progress. I decided upon using Houdini for my workflow since my earlier research suggested that it is the primary and most versatile software package that is readily available. Lots of companies and studios have in-house solutions, whilst lots of others simply use custom shells of Houdini. The initial step for this stage was to gather research and reference materials that would help guide me.
Figure 9: Gathered reference images for conducting a digital groom in Houdini
Figure 9 highlights the references I used for this practical. I tried to find a good mix of close-ups of certain body areas as well as wide shots to be used as an establishing basis. Alongside these, I gathered various video guides and documentation on the grooming process in Houdini to teach me the fundamental workflow (SideFX 2025; Esfehani 2024). Since this was the initial groom on top of which the others were based, I spent additional time simply researching and watching guides for various methods
Figure 10: Demonstration of Practical 1: Generating a Clean Guide Groom
Once I felt like I had enough supporting material, I started on the guide groom creation. Figure 10 highlights the workflow and pipeline I used within Houdini, starting by importing the wolf model I had previously sculpted. A groom can only be as good as the underlying geometry, and if it's not anatomically correct, then it can drastically reduce the level of photorealism. Therefore, when sculpting the wolf, I made sure to use anatomic orthographic references to help align the model and get as accurate a representation as possible. The next step was to start generating guide hairs on the imported skin and separate it into different parts of the body. The fur on the mane is rather different compared to the fur on the legs and face, so I meant I had to use a different groom pipeline for each area. As the demonstration shows, I did this by using some basic VEX code in an Attribute wrangle node to create mask groups and paint the attribute in specific areas.
From here, the remainder of the time was spent on artistically grooming the wolf and making small creative adjustments. The core pipeline I used was setting up a guide groom node for each body area, which then feeds into an individual HairGen node. I kept adjusting different areas and repainting the masks as I went along to get a better blend between the different fur layers and to create additional separation groups as necessary. Within the guide groom, you can control the direction, frizz, length and hair flow using guide process nodes. These allowed me to artistically drive the groom in a procedural and non-destructive workflow. The main purpose of the masks was actually to determine different density values, which were determined within the guide nodes themselves, but also by the HairGen for the density of the interpolated hairs. This workflow got me roughly 90% of the way there, with the last 10% being driven by a manual sculpt of the guide hairs. This is where I spent the majority of the time, fine-tuning and constantly referring back to the reference materials as I went. The final renders were completed using Arnold for Houdini, since I found that to be the least noisy option, compared to Mantra and Karma.
Evaluation
Figure 11: Render results from practical 1
The results of this practical were evaluated in conjunction with Practicals Two and Three, since I am investigating the audience perception between the three. From an evaluation standpoint, this practice aimed at creating a technically perfect and ‘clean groom’ based on the reference material, to act as a comparison point. This also means that it is the least computationally intensive practical since it just contains the groom itself on a static mesh. I knew that I would only need a frame for the output, as I wasn’t assessing animation, simply a static groom. This meant that I didn’t have to deform the hair to animated geometry, which would’ve resulted in much harsher processing requirements.
Figure 12: Casual audience member responses on the realism of a clean groom
Figure 13: Academic/Professional responses on the realism of a clean groom
The feedback gathered for this was based on a Likert-style response form ranging from Not realistic at all to highly realistic. As seen in Figures 12 and 13, the clean groom consistently scored lower than its more developed counterparts, particularly with casual audience members. Audience members gave it an average of being slightly realistic compared to the real-world reference images they were provided with. This is likely as a direct result of the procedural tools used to control the hair flow and clumping direction. These often have the result in displaying as unnatural and rather linear deformations due to the procedural nature in which they are generated. This is where a guide groom node would come in for manual grooming and creating some guide break-up and clumping. This was clearly outlined in the qualitative feedback, seen in Appendix A, with one respondent saying that “the fur sometimes looks not natural in the way it flows” and another thinking that “Lighting and overall clumping of the hairs could be more accurate”.* This highlights that the perceived realism of a clean groom is diminished without added break-up and fur clumping when in a natural outdoor scene, like the one showcased. However, when viewing Figure 11, some respondents also felt that the groom “looks near perfect” and that the “hair looks great”, but that they felt the face threw them out of it.* This actually highlights a key point that I found from my earlier research: A digital groom can only be as good as the underlying anatomy. When it comes to real animals, audiences already have a preconceived notion of what they should look like, meaning that even the smallest inaccuracies can take them out of the scene. This means that whilst the procedural workflow I was using was efficient and provided the intended result, a more hybrid workflow could improve the realism. Details like added clumping, improved face topology and more natural manual grooming of the hair flow would have greatly improved the perceived photorealism of the shot. Technically accurate grooms rarely exist in a real-world scenario, which likely limited the audience's perception of this practical.
Implications for Practical Two
These potential limitations were also identified in my preliminary research, where artists often prefer manual grooming methods for added creative control, but also to create more natural-looking breakup (Esfehani 2024). This means that, alongside the added environmental simulations, I should more carefully define hair flow using manual brushes and create added clumps and breakup within the HairGen SOP. This should allow for a better blend across the various grooms and a more fluent transition. Additionally, the groom would benefit from improvements in the facial topology as well as the eyes, as they drew many participants out of the scene. However, for the sake of this investigation, the underlying mesh will remain the same to ensure consistency across the practicals, since I am aiming to compare the effect of environmental imperfections rather than anatomical accuracy. Changing the topology could result in a more favourable result for the remaining practicals, even if the groom is of a diminished quality.
Practical Two
The second practical aims to add an initial layer of imperfections that would be believable in the given scene. For this, I chose to add water and rain to the scene, as well as a developed groom based on those parameters. Given the initial response and preliminary research, I also aimed to add manual grooming methods to aid the simulation if needed.
Figure 14: Demonstration of Practical 2: Using dynamic fluid simulations to create groom imperfections
The previous demonstration was a foundation on which to build further, meaning that I already had an underlying base to work with. Figure 14 showcases the workflow and steps that I used to achieve a ‘wet fur’ effect on my groom from Practical One. This part of the process was much more computationally intensive since I was creating a dynamic simulation, which had to be cached out prior to working on the groom itself. The FLIP fluid sim, which I was using to drive the fur, was able to be rather simplistic since I only required it to initialise the velocity and to form as a basis for the creation of a Wet Map. This is a common practice in Houdini, which I discovered during my research for the initial practical, with Wet Maps offering a versatile range of applications, especially when creating wet fur (Stavginski 2017). Groom nodes in Houdini are basically just large VEX containers that use the said language for all computational aspects, which means that using attribute wrangles with VEX expressions was necessary for certain parts of the workflow. This is especially true when creating interactions between different systems within Houdini, such as DOP and FLIP fluid simulations. For the Wet Maps, I used VEX expressions to retrieve velocity and collision attributes from my FLIP simulation and feed them into a solver using my created UDIM tiles. When imported into a COPnet, this outputted a 2D grayscale mask, which was animated on a per-frame basis using the FLIP fluid.
This Wet Map formed the crux of the rest of the practical, allowing me to create guide masks based on the texture to control the various groom attributes. The initial approach was to use a guide simulation node with the FLIP velocity as the basis for the stiffness, clumping, and damping parameters. Based on my research, my approach was to utilise a wiresolver in an external DOP network and use a vector field dynamics node to bring in my attributes. However, it was at this point that I couldn’t get the wiresolver to represent the guide curves accurately, breaking any simulation I was trying to run. In Houdini 21, the updated and more current workflow is to use Vellum solvers for any kind of hair simulations, though I couldn't locate any updated documentation or guides on how to bring in custom velocity attributes in the same way. My wiresolver method was based on a guide working in Houdini 16.5, which had a working but now-deprecated workflow. Several updates and changes have been made to the solvers between versions, causing me to research alternative methods. Without a further deep dive into Houdini attributes and creating custom Vellum hair simulations, I would not be able to progress further with this pipeline. Pursuing this would outgrow the allotted time and the scope of this investigation, and therefore, I reconsidered a more manual approach. Instead, I use the created guide mask and Wet Map texture to separate the groom into different hair-generation nodes, and a manual guide groom brush to adjust the parameters. Given that Practical One found diminished results due to a procedural methodology, I opted for increased clumping and frizzing here with a manual groom, determining affected areas.
Evaluation
This practical was much more divisive, both in terms of the attempted methodologies and the feedback responses. My main methodology was well-founded, with various sources backing up the use and creation of wet map textures to drive shading attributes (InsideTheMind 2024). However, my methodology for using a wiresolver for the groom simulation was outdated, and investing more time in the updated Vellum solvers would have greatly improved the efficiency of my workflow and helped meet the original scope set for the practical. Using the generated Wet Map textures in a manual-based groom was a good alternative, since it gave me greater control over the visuals.
Figure 15: Render results from Practical Two
The visual quality and integrated aspect of the groom in this practical, seen in Figure 15, benefited greatly from the use of Wet Maps. They allowed for far greater control during shading and provided a visually realistic basis, since they were generated from a physical water simulation. This solved a key issue in Practical One: the technically accurate, clean groom would never be found in a natural environment. Whilst they are procedurally generated, which makes them very flexible and reusable across various aspects, such as shading and guide masking, they also act as a good blending point into manual control. The decision to keep the FLIP simulation relatively simple was also an aid in this, since it meant I could easily adjust the interaction feeding the mask without wasting unnecessary computational time. The critical limitations in this practical came from the use of VEX-based operations and an outdated wire solver workflow. VEX is a powerful tool that enables complex operations and interactions, but it has a steep learning curve and can be error-prone and time-intensive, as I established. If I had identified this in my preliminary research, I would have had more time to investigate ways to implement the code more efficiently for my desired purpose. Furthermore, the outdated workflow I initially used was a clear limitation of the methodology, as it no longer reflects the industry-standard pipeline, which had transitioned to Vellum in Houdini 21. The initial reasoning behind this was that I could not find an updated documentation for the specific effect I was trying to accomplish. A more comprehensive look at attributes and Vellum solvers in Houdini 21 would have allowed me to adjust the workflow to my needs. This would have enabled a visually photoreal groom, which also had physically accurate attributes, improving the overall cohesion.
Figure 16: Perception of realism for Practical 2 by Casual Audiences
Figure 17: Perception of realism for Practical 2 by Academic and Professional Audiences
Audience feedback reflected this as well, though it proved more divisive among Casual and Academic audiences, as seen in Figures 16 and 17. Casual audience members found it easier to be taken out of the scene, with the majority of respondents leaning towards the groom being unrealistic. However, respondents who were more familiar with VFX and grooming workflows gave it a much higher score, with ~90% deeming it slightly realistic or better. This potentially stems from causal audiences solely relying on the images provided, as well as their own perception of what a wolf should look like. The qualitative feedback was also very divisive among Casual audiences, with some claiming that “the fur should interact more naturally” with others claiming that the “darker and more scraggly with thinner” was more accurate with “my version of a wolf”.* Additionally, audiences were increasingly commenting on the topology, particularly the face, and felt the eyes “look unnatural”, “were really big”, and “didn’t look quite right”.* This demonstrates that the fur clinging to the underlying topology drew their focus to other areas of the model rather than the groom itself. This inadvertently shows that it seems to be more realistic, as audiences had more to say about the overall scene rather than the individual groom, as the two were better blended. In contrast, academic respondents found the scene to be more realistic compared to the practical one and were also able to provide more critical feedback and identify certain areas of the groom which drew their attention. Academic respondents identified that “the groom with a gradient is the better of the two, representing the real-world references” and that the fur had “realistic weight and direction”.* This shows a clear division, where respondents familiar with Visual Effects workflows were able to isolate the groom and likely compared it to their own work or realistic expectations from a digital groom. Respondents here also identified the face as a focal point for improvement, citing the need for a “shorter and more detailed” facial groom compared to the rest of the body.* This reaffirms my findings from Practical One that the topology is as important to the realism of a groom as the fur itself. Though this was a controlled variable during this investigation, and therefore was not going to be adjusted between Practicals.
Implications for Practical Three
This practical highlighted a few clear improvements moving into Practical Three, particularly regarding the environment and overall integration into the scene. A few respondents mentioned that the environment didn’t always fit the scene and that it should “emphasise a more natural pose and surrounding”.* This tied nicely into Practical Three, as I’m generating resting and falling snow interactions, which integrated better with the snowy HDRI environment I had been using up until this point. The HDRI was a deliberate initial choice since I meant to replicate an Arctic Wolf, which is often found in snowy conditions. A dry forest background could’ve improved cohesion for Practicals One and Two, but this was a controlled variable to maintain consistency and comparison points throughout the investigation. The manual grooming methods here proved effective in creating a visual representation of what audiences expected, meaning I will be maintaining that through to Practical Three, though adjusting for the new parameters. The overall results of this practical suggest that as environmental imperfections are introduced into a groom, the perceived level of realism increases, though it also increasingly relies on anatomically accurate representation and smooth integration with the environment and surroundings. This also highlighted that audience perception can vary greatly across familiarity with the content materials, which I expect to continue into Practical Three.
Practical three
The final practical focused on building a falling-snow simulation using a DOP Network and generating some resting snow on the base groom. I would also be concepting, incorporating Practicals Two and Three into a single render for added environmental effects.
Figure 18: Demonstration of Practical 3: Generating falling and resting snow using a DOP Network in Houdini
As seen in Figure 18, I initially built directly off Practical One and started by generating a standard POP Network (DOP particle system) to simulate the falling snow. Real snow is very light, meaning that it’s easily influenced by small changes in wind direction, speed and humidity. I replicated these effects in Houdini using a variety of POP Force and Wind attributes to introduce air resistance and turbulence wind forces to break up the simulation. In terms of the node graph, this was a relatively simplistic setup, with the main quality driving points coming from attaching individual frequency curves to each parameter and adjusting them whilst looking through the camera point of view. I kept the initial particle birth rate low to keep the simulation computationally light while I was still adjusting the parameters, and then increased the sample size to produce a final high-quality cache. The initial plane I used for a source also had a frequency amplifier driving the offset for the mountain node, meaning that I could duplicate for midground and background snow offsets, at a reduced birth rate, without risking uniformity. For the snow geometry itself, I tested around with a few versions, including a low-poly model of a snowflake for added realism, though I decided upon a simple low-poly sphere for the sake of saving processing power. Initial individual caches I was doing with the snowflake geometry were taking upwards of 5 hours apiece, which was too expensive for the time and scope of this investigation. Additionally, in the final render, the quality differences were marginal due to the added motion blur and velocity blur. The skin mesh and hair geometry were added as colliders within the POP network so that the groom would appear to be slowly collecting snow particles over time. This was then also aided by a manual project to simulate snow particles using the Ray node to recreate the resting snow after a heavy snowstorm and to replicate the snow reference images I had gathered in Figure 9. The results of the Ray were converted into a VDB and then back into polygons to create a smoother mesh with improved topology, improving render quality and speed.
Evaluation
Figure 19: Render results from Practical 3: Falling and Resting snow from a DOP network
The initial results I gathered from Practical Three were very mixed, which I found was primarily dependent on the Melanin value as well as the lighting conditions. As I established in my research, Melanin is the substance that causes hair, eye, and skin pigmentation in humans, as well as in wolves and other animals (Cleveland Clinic 2022). For fur, it determines not only a base colour but also specularity. Naturally, different colours absorb different amounts of visible light, which makes a Melanin value closer to 0 reflect much more light compared to a value closer to 1. Snow is also a very reflective material, meaning I had to carefully balance the two as well as the light intensity so one wasn’t overshadowed by the other. Additionally, my initial process of purely using a DOP network for the resting was sound in theory, as the particles would gradually gather on the surface as the simulation compiled. However, as stated previously, initial caches were taking upward of 5 hours, at which point not enough points had gathered to create a realistic render which demonstrated the snow effectively. Therefore, I used the Ray node to project particles onto the model, resting on the first points they came into contact with. This allowed me to have a physically accurate representation of rested snow whilst also keeping the scope of the Practical in check and not requiring 10-hour simulation caches.
Figure 20: Perception of realism for Practical 3 by Casual Audiences
Figure 21: Perception of realism for Practical 3 by Academic and Professional Audiences
Practical 3 proved to be equally divisive among Casual and Academic audiences, with the latter favouring this as the most realistic compared to the other two Practicals. As seen in Figure 20, casual audiences felt that the final groom was generally realistic, though not perfect. Different reasons were given for this; some respondents felt that “the snow is not resting properly on the fur” and that the “snow placement does not look very realistic”, whilst another felt that the final groom was the best and that “the snow sat on the fur and the face looks exceptionally realistic”.* This was in stark contrast to Academic respondents, ~80% of which found the groom to be highly realistic which was a definitive improvement over Practical One and Two. Respondents here felt that the “snow placement and its interaction with the fur is good” and the “snow on the fur makes sense”.* This shows the perceived level of realism for this groom greatly depends on the audience's level of expertise and reference for what resting snow should look like. The measurable improvement in the perception of this Practical also reaffirms that topology and integration into the environment have a significant impact on photorealism, as the snow fits better into the snow HDRI environment used for the investigation. Though ultimately, the audience's perception of this practical hinged on the quality of the snow itself. The methodology I used for the now caused it to appear slightly more liquid and sharp compared to the reference material. This was reflected in some of the academic responses, which highlighted that the snow looks “too vicious with strong edges”, breaking some of the immersion.* For further visual fidelity, I would incorporate some of the colliding particles as flakes and reduce the smoothing of the resting mesh to provide more accurate results. Overall, this practical demonstrated that adding environmental imperfections does have a measurable effect on the perception of the realism for a groom, but it highly depends on other factors like underlying topology and integration into the scene.
Conclusion
Overall, it can be confidently summarised that environmental imperfections increase perceived realism of a digital groom, though this hinges on a few key factors. Firstly, the overall execution must meet industry standards and the level of photorealism expected by a general audience. Casual audiences tend to review the scene as a whole rather than a single aspect of the shoot, which puts emphasis on the overall integration and blending between various elements. Secondly, for a digital groom to be considered photorealistic, the underlying topology and anatomy have to be accurate and representative of real-world references, especially when re-creating real animals. This was evidenced by the progression in the perception across the Practicals, with a marked improvement in overall photorealism. Practical One showcased a procedurally clean groom, which was technically accurate but lacked artists and visual direction, throwing audiences out of the scene. Practical Two built on this by introducing wetness and damp clumping to break up the groom and provide some variation. It showed a measurable improvement in quality and overall perception, though it was hindered by a partially physically accurate simulation and by the environmental integration not fully aligning. This was backed up by the initial research, which also found that the anatomical accuracy is highly important when creating a digital creature since it is the foundation upon which everything else is built. Practical Three proved to be the most successful, with audiences having a stronger reaction to the snowy integration between the foreground elements and the environment. This was reflective in the methodology used, which was a hybrid blend between manual and procedural workflows, which is the industry standard for creature grooming. Audience perception of realism does not stem from technical accuracy or the workflow used, but from the visual fidelity and integration. However, this investigation and grooms in general are greatly aided by physically accurate simulations, which are controlled by carefully maintained and manually driven parameters. This proves why the advancements of AI and procedural workflows will have diminished results since the last 10% of a groom will always fall to the manual artist, with the greatest impact on audience perceptions.