Saltstraumen Maelstrom Process Blog

11/9/22 Update

Here is a short render with the updated water. My computer restarted midway through the render, so this update is only 3 seconds since the rendering was cut short. 

11/8/22 Update

Adjustments were made to the white water. They are subtle, but feel a little more natural compared to the last version. 

The look of the water is being tweaked based on feedback and correcting a few issues such as the visible line between the ocean and flip tank. This is a test render of a frame of the water. It is caching and will be rendering out tonight. 

11/5/22 Update

First render with white water. The pre-roll of the water sim may need extended to avoid that burst of white water and the white water needs to be expanded. Right now it feels unrealistically limited to the whirlpools.

11/1/22 Update

Rendertimes are back into acceptable ranges. Was able to get a render test done for the simulation. Whitewater is currently being simulated and cached, I hope to have a video with those elements soon.

10/31/22 Update

I have managed to get a working Redshift setup, however there were challenges with the texturing and displacements. The mantra ocean material was allowing some of the subsurface details of the simulation to be faked, but Redshift does not have a comparable shader that I have found. In addition, geometry between the flip tank and the flatten padding creates a wall underneath the water that I have not been able to get rid of, however, the process of getting Redshift working with the spectrum may have led me to a Mantra rendering solution.

The Ocean Spectrum displacement map during render was the source of the increased rendertimes. Since Redshift was not working with the rendertime displacement, I found a solution with the ocean evaluation node to physically displace the geometry before render. At first this seemed problematic because it required me to add a lot of division to the flattened geometry, but actually provided me a way to more precisely control the geometry needed for the ocean.

I have ran a test with the Redshift and Mantra setups to see how they perform with this setup. Redshift is using a single light source, Mantra has the three light setup, however both are using the same geometry generated from the ocean evaluation node.

Redshift also appears to need the height field converted to polygons to render properly, even for creating the landmass mattes.

Render time: 2:31

The mantra render has lost a little of the diffuse map refinement that was applied with the render-time replacement, but overall is giving me the best looking result so far. Tonight I will refine the ocean spectra and let it cache. Once finished, tomorrow I will work through the look development of the water and then run a render overnight.

10/30/22 Updates

It has been a difficult week trying to solve my segmentation fault error. The issues would crash Houdini on the first frame whenever I attempted to run a high resolution simulation. After a lot of troubleshooting, it appears to be a version issue with Houdini. I am now working on the 19.5.403 version.

Above is the current simulation, it was 240 frames at .025 particle seperation which took 24 hours to simulate and cache the mesh. In addition, I have been trying to optimize the scene for rendering and improving the materials. However, I am not sure the materials are working quite right with the height fields.

The next major issue to solve is render time. The scene has been split into various render layers to make the rendering more efficient, but the water render is currently clocking in at around 40 minutes per frame. Some of this appears to have to do with the Ocean Spectrum displacements. I am testing a workflow using Redshift for the water rendering, but it is proving to be a more significant change to the setup than anticipated.

10/23/22 Updates

I fell behind on updating the blog this weekend.

This weekend I separated the various scene elements into different render layers to better manage my time when it comes to finally rendering the scene. The layers are below, with the exception of the atmosphere. The atmospheric perspective will be moving to a post compositing method using either Nuke or After Effects.

Overall the scene scale has increased some to better reflect the real-world proportions of the river and add some more distance between the effects and the shore line as well as the background landmass from the camera. This has increased simulation times somewhat as there are more particles being generated.

Notably, the surface meshing timescale has exponentially increased with the larger ocean padding that is required for the ocean spectra. The current simulation is rather low res for the sake of time. 

Background/Midground Terrain

Foreground Terrain

Water

Trees and Rocks

I am experimenting with rendering the white water as particles rather than volumes. The results seem more defined and promising than the volume approach.

Tests were conducted in a separate test file to attempt to get more detail from the whirlpool velocity fields. Multiple variations were attempted, but the curve and the resulting whirlpool shown here were the best I was able to achieve. However, the river velocity is reducing the details seen in the simulation. 

Goals:

• Refine White Water and Whirlpools
  • Run a hi-res simulation created to check water detail
  • Adjust the white water emissions and look dev to be more organic
  • Add motion blur and/or create hybrid volume and particle setup
  • Increase whirlpool size
• Correct Ocean Spectra
  • Correct the chop vs calmer waters
  • Adjust whitewater so it is actually visible
• Test compositing workflow with atmospher, including producing depth map
• Produce at least of a flip book of white water and water or a test render.

10/18/22 Update

I am refining the whirlpool velocity fields and started introducing the whitewater setup. There is still a lot of refinement to work through, but the shot is starting to come together.

Goals this weekend:

• Continued refinement and improvement to whirlpools and ocean. Specifically adding in more detail to the whirlpools and modifying the simulations to have a better sense of scale. Whirlpools may be too large/close to the foreground shore.
• Whitewater refinement: Iterating over the simulation sources, emissions, and material to achieved better definition and more accurate movement.
• Adjustments to the landscape. The rope fence seems too small and the scattered rocks need material detail to show some snow cover to integrate better with the environment.

10/17/22 Update

I have worked through some of the issues. The masking for the Ocean Spectrum textures has been corrected and is not applying to the calm areas of the FLIP fluid simulation. There is a slight line at the boundary of the tank, which I think can be fixed with some adjustments to the flatten geometry settings.

The current whirlpool simulation is too strong. With the recent changes, the settings were disrupted. This week will focus on refining and fixing the FLIP simulation and fully integrating the white water simulation.

The ocean spectrum is also throwing off the scene scale slightly. This will be addressed once the simulation is more refined.

Render time: 10:41 per frame, local render.

Main Mask

Currently working on:

• Integrating the FLIP fluid simulation with an Ocean Spectrum setup
• Refining textures and lighting
• Tweaking whirlpool settings and sourcing

Goals this week:

• Major progress on white water simulation
• Continued refinement on water simulation
• Cleaner integration between FLIP and Ocean Spectrum

Current Settings:

• Frames: 72
• Particle Separation: 0.02
• Particle Radius: 1.2
• Water simulation time: 9 hours, 27 minutes
• Surface Cache Time (Separation bumped to 0.04): 45 minutes
  • At the 0.02 separation value, it completed in 6 frames in 35 minutes

A great deal of testing has gone into shading methods with the height fields. Current difficulties with the height field shading and render stem from detail limitation in the height field mesh. In an attempt to keep the scene as efficient as possible, I am avoiding converting the height fields from volumes to polygonal meshes. The Polygon meshes produced from the height fields are very heavy in the scene file and would require an alternative texturing method as the HF mask falloffs are not respected.

To approach the shading of the height fields, I am working with a layered method in the mantra material builder (thank you Professor Folwer for this technique). Using a material builder allows me to bind any masks created with the height field which enables me to mix various material together. 

Typically my workflow involves separating the scene elements by anticipated render groups for compositing. Within each geo node for the scene setup, group nodes are used to isolate the various elements in which to attach materials. This workflow reduces the amount of material nodes, making it easier to adapt and change if needed.

The height fields have introduced complications in this work flow. Traditional merge methods for the landmasses result in only one of the three terrains showing up in rendering or as collision geometry. I have not found a good method within the height field tools to combine the terrains into a single entity. To work around these issues, I have merge the individual terrains into their own respective geo groups and applied a material node to each height field. It is a minor inconvenience, but the work around is working fine for this scene.

Ocean Spectrum Sourcing – Multiple ocean nodes creates more interesting results

There are three major areas of improvement for the ocean integration for me to move forward with the technique:

• Low activity areas of the FLIP simulation need to have the ocean waves added in order to be convincing/realistic. This may require adjusting the flat tank to a guided tank simulation setup.
• The boundaries of the FLIP fluid have been unpredictable. Often times creating noisy or jagged edges where the ocean and FLIP fluid meet.
• The Ocean Spectrum setup needs to be refined to give a better sense of depth with an appropriate level of wave and chop. Currently, the setup seems finicky. Changes to the values either cause the waves to barely show up, or result in too large of an effect. My lack of experience working with the Ocean Spectrum setup is likely the cause of this issue.

The latest iteration appears to show that the mask will need refinement for different resolutions. The previous simulation used the same settings, but the mask was much more pronounced.

This most recent simulations particle count was quite large, and the surfacing node had to be adjusted to make the caching times reasonable. As such, this version lost details during the meshing phase.

Early Render Test

Current Render Test

I have dedicated the better part of a day or two to reworking the landscape geometry to try and achieve a more realistic and interesting landscape. The results shown here are created using height fields, copying to scattered points, and procedural modeling for the rope barrier.

The landscape is close to the desired final look, but will require refinement through additional details and texturing. Most of the refinement will focus on the foreground elements as the plan is to add some atmospheric perspective to de-emphasize the background landmass and add to the scene’s mood.

The terrain of the scene is generated using height fields. To create a fairly accurate landscape, a topology map was downloaded for the Saltstraumen area and brought in as the initial data for the height fields.

Additional procedural sculpting of the terrain was needed due detail limitations of the original topology map. To ensure the height fields are as efficient as possible, the foreground, midground, and background were split from the initial setup and handled separately.

While learning about height fields and working through some of the challenges of making this landscape, it became clear that getting realistic detail in the landscape requires a ridiculous amount of voxels. Although voxel data is typically fast to calculate, the height fields quickly became cumbersome with the level of detail I was trying to capture.

By separating the different regions of height fields, I was able to handle the resolution requirements more specifically based on how close to the camera each particular land mass is to the camera. While I am fairly happy with the terrain results, textures and additional geometry details needed to be added to sell the realism of the scene.

Foreground Height Field

Midground Height Field

High resolution render geometry

High resolution geometry setup for rendering

Low resolution height field setup. These severely reduced volumes will be used for collision geometry in the FLIP simulation.

Note: the current height fields as collision objects did negatively impact simulation times. I may need to reduce the detail further, but have concerns it will reduce the accuracy of the collisions. 

Snow Mask

10/10/22 Updates

Started an initial pass on texturing. I have been looking at a snowy setting, so I used the heightfield masks to place different textures for the stone and snow. This approach, so far, has some drawbacks. The masks created by the heightfield look fine in viewport, but when trying to render the water disappears if the heightfield is left as a heightfield voxel.

Research into texturing terrains showed many different techniques, but I started with just converting the heightfield to polygons and trying to work with the groups created. The result is not refined enough, and the point groups lose the falloff from the masks, resulting in jagged edges to the snow. The materials I am currently using is throwing off the scale and the results look fake at the moment. This is especially true in the foreground.

The next iteration of the textures, I will be exploring exporting the height, flow, and mask data from the height field and using those maps in Substance Painter to mask the different regions needed for the materials. I believe this will give me the best control and will preserve the gradation of the masks.

Lighting is just an HDRI at the moment. The light and shadow patterns are too flat and the color is too warm compared to the reference.

White water will need more time to get under controls, this current version is driven only by vorticity and curvature. The white water simulation will need time to refine, but I felt it was best to start adding these effects now so they are being refined with the water simulation.

Additionally, the sense of scale with the water is off. The waves in the background are too large and makes the river look too narrow. 

Terrain setup. Due to the heaviness of the height field geometry in the file, an height field output node is used to capture the geological features and then loaded from disk. A fork of the terrain is reduced and converted to volumes for the FLIP Fluid collision detection. Tests will be conducted to see if the volume or polygon soup method works best for the collision. 

Reference image.

10/3/22 Updates

Quick update this morning. Did a test with the landscape collider, and the wide gap of the river meant a large increase to the simulation tank size. There are issues with the collider that need to be fixed, I will be looking at alternative ways to fill in the river to make it more efficient. Camera angle is in progress as well, right now it does not represent the whirlpools well.

The frame count is really within the run-up time, so the initial moments of the whirlpools forming will not be seen in the final imagery.

10/2/22 Updates

Moving forward with the volume velocity from curve method of creating the maelstrom, I have been testing how different curves might affect the various looks of the maelstroms. The tests are below with their corresponding curve shapes. Very short simulation times are used to keep the iteration process manageable. 

Generated Height Map

Basic fluid shader