水出しカフェ [Mizudashi Café]

This is the blog for the MA Games Art and Design at the University of Hertfordshire 2021-2022. There is a navigation bar to the right (press the ‘navigation’ black button to unfold it) but here are the links to the most important categories either way!

Practice 1

Development Projects

Weekly projects

Practice 2

Practice 2 posts

Live Brief

Visual Spirited Away

Studies

Final Project

Spirited Away

Final Project

Learning Process Documentation

Research & Media

Annotated Bibliography

Reflective Posts

Critical Analysis

Other links

Art Station Nepre

Itch.io

Weekly projects indexed by topic

Red vs Blue

Places of Mind

Nostalgia

Light & Shadow

Academically I believe this project has resulted very positively. I have tried to focus on what was required, both with the documentation and with the planning folder. Most of the project deliverables have been accomplished, if not with brief changes, and some of the stretch goals have been met too. For the promotional video I have decided to create two versions: a voice over one, where people can meet me more, knowing the intentions of the project and myself as an artist, and a second one where strictly the footage with the music is shown. I believe this second one will be better for general audience, such as Studio Ghibli fans, while the first one will be more important for employers.

In terms of the documentation, having a daily output of content was easier to follow than just having the stretch goals originally planned, as some of the notes I took would be merged into fewer posts and resulted in less detail on the final posts. I have divided it into three categories, the first one which has everything to do with Spirited Away. This contains posts from the pre-production to the final result and usually has a lot in common with the second one: Final_Project. This category is where everything related to the project itself, where sometimes things unrelated to the series were written. And finally the documentation, where this post exists. This is where a more intimate conversation happens about my creative and learning processes.

For the presentation I tried to speak a bit more about the technical and conceptual challenges of the project. It has not always been easy, but I believe it is there where you can learn the most. This have perhaps leaned into a bit too technical aspects perhaps, that is why I reserved it for the professors instead of general audience. Although if anyone is interested I have redirected them to this blog.

Finally for maximum reachability I am posting this project on many social medias. The more common ones for the general audiences: Instagram, Twitter, Reddit and YouTube; and with a bit of more detail onto Artstation and LinkedIn for employers and likeminded professionals. It has had so far better repercussion than the Bioshock project within Reddit, but worse on YouTube. Although these projects grow slowly on YouTube, while on other socials they grow quicker and then go to die as fast as they grew.

The final piece of this project was rendering. Thanks to the newly added - 4.26 - render queue, setting the shots and settings with a more confortable UI was really easy. After asking for feedback, I noticed that the main building was too contrasting on the more distant shot, so I decided to make a couple of close ups.

A version without voiceover can be found here:

This last stretch I have been working on the clouds and a bit of layout. I have been following this tutorial about Ghibli clouds and it has been really interesting. He uses noises and normal maps as a way to work with the light within the renderer. 

The main problem that we have, similarly to the trees, is that Unreal Engine doesn’t have many of the features that Eevee does. For example, the color ramp is non-existent. We could have done as we did with the trees and take them to Unreal Engine and then hand make a colour ramp but... I thought it would be a better idea just to render the clouds with a transparent background and take them to UE. 

This way the are planes, they don’t need to be 3D. Cheaper, and give a sense of scale, as really distant clouds don’t move when you do. 

It makes a really big difference, I recommend checking out the previous posts and compare the image with and without clouds!

Today we are coming back to the post-process material we had, and I wanted to have a look at masking. Right now the outline is applied to everything, including the skybox, and although there is a manual way to mask these things, I believe automating it would be best.

First of all, we would have to enable the custom depth stencil. This will allow us to select specific objects and make them render on a different layer. We will be able to get this information then and use it to mask our post-processing.

After this, we can get it into our post-process material with the scene texture -> custom stencil.

To get the information, we can just go to the mesh and tick the Render CustomDepth Pass checkbox and set our value to a number. As we want a true or false - these can range from 0 to 255 - I just selected 1.

The problem with this approach is that it would apply to the whole element, not taking into consideration any occlusion.

To solve this, I found this clever solution online. Using a custom code node, instead of the Scene Texture stencil, BYC writes this code:

First, we make check for the Scene Textures. After that, we gather the Screen Space Data from the viewport with the Screen Position UVs, and then subtract to the custom depth that we got the depth data, ceiling and clamping this afterwards. These are made to get integers rounded up and set between 0 and 1. With this information, we multiply by an alpha that we set, to lower the values.

Now we can see that the depth is taken into consideration when calculating the values of the custom stencil, and allows for a much better look.

As this kind of sharp lines doesn't suit very well the environment, there is an opacity value that Lerps between the normal screen values and the post process values depending on the range that we set. This way we can have a more subtle look to it.

This is how it looked before any post process. It is a very subtle change, but I believe it adds to it, and I got to learn a lot from this experience.

In order to follow the concept we talked about previously, I had to make some changes though. 

If you look at the first implementation of this concept we can see that although similar, it lacked some depth to it.

The colours used by Diana show why it looks flat:

There is some variation depending on the distance to the camera. I was a bit stuck here, as I wasn’t able to hand paint the floor without losing the grass. For some reason it didn’t allow me to add other grass outputs in the Landscape material for the different layers. This was a bit dump on my part, as you can just add it to the existing one. This way I painted some colour variation, and the results are very different in my opinion.

The previous layout seemed a bit flat. I hadn’t considered how it seemed before, so I had just been creating the tech to support the scene.

Old layout.

After talking with some friends, they recommended that I chose some other views, even doing some concepts for me. I am very lucky to be honest, as they have been of great help.

The main one I have followed is the next one. It gives a sense of direction and perspective, while filling the areas that would remain empty.

And although I have reduced a bit the distance, I must admit that it looks significantly better.

Though there are some alternative views that I would love to implement and that I believe would be great too. But I depend on creating the clouds still.

Something that has also changed has been the shading on the Trees and Rocks. Before, the trees had the fake highlights + a lit material, while the rocks had full on texturing.

This has changed recently for the approach of all unlit + fake highlights. It works wonders as it is on the trees, but the rocks required a bit more work, specifically with the transition, which was done by applying a texture to the threshold, so it looked a bit more rough.

It results on a better look and more importantly closer to the references

These days I have been working on the main building. As I have mostly used techniques already explained I thought there was little to write about, but today, while creating the main walls, I realised that I had to do something else.

The approach I have taken is a bit different than the usual UVs -> Substance Painter -> Unreal Engine, but rather, after doing the UVs, I created a material in Susbtance Designer that would serve as a base. Ignore the colours as they haven’t been tweaked yet.

This tileable material would be good enough for me to just apply it to the UVs, but some would look a bit weird, so by baking a mask with the curvature map and some painting in Substance Painter, I can apply a darker shade onto the building on itself.

This helps everything blend together a bit better, while maintaining the quality of a high-resolution texture. As the mask only is applied as an on-off.

These days I have been working on a set of tools that would help me achieve the look I want on the environment.

How my spline tools work, the base comes from this tutorial, is by spawning an object per each spline point, excepting the last one, with the data of the position and tangent of the spline points themselves.

My personal touch comes on setting variability on those settings, so that it is not strictly on the points themselves but on random values that are nearby them. These values are fully customizable in editor.

And example of it working can be seen on this gif.

On top of the simple-grass-wind node from unreal, I wanted to add this animated element where the grass moves and shines with the different wind gusts. To approach this, I simply made a texture on substance designer that would pan on top of the absolute world location of each blade. When there was white, I would increase its brightness and move them a bit quicker. On top of this, there is an element of noise that would hide and show different parts of the wind with a perlin noise mask.

The panner material is quite simple, but I will add it anyways.

For the trees, what it is usually done is using a technique where a low poly element is set up underneath, and on top of it some particles that act as leaves - following the normal vectors - are set up.

And although I really like this look, I thought it wouldn't be the most suitable for the look I wanted, so I followed a tutorial by Lighting Boy Studio, where he explains how he made this effect in Blender.

Though very cool, we are in Unreal currently, so I wasn't sure how it was going to turn out in the end. I followed everything up until the shading, as Blender has some elements that Unreal does not - Color Ramp in this case. When taken to Unreal Engine and set up a basic colour, they already looked good enough, but to add a cherry on top of it I created a fake shaded material which allows me to fake some highlights. I knew more or less how to approach this, but this tutorial was of great help.

This differs a lot from other types of Cel Shading. Usually it works on the post-process level, making it dependant on the angle of the camera, here as it is fake it just need the light angle - 45 degrees pointing to the negative X - and that would be enough.

With Highlights

Without Highlights

This was an important change. Before, the texture was bigger and softer, this way we have got more patches, similar to how they usually are in Ghibli Movies. My intention is to create a more patchier look when it is close to the camera, and fade into these clumps when they are more distant.

To achieve this look there were numerous iterations. As depending on the gradient that you use, they might be noisier or softer. And despite being the same bnw noise, the results vary enormously.

If we compare it to the previous image, we can see that the colours are way more vibrant.

This is to get closer to how Ghibli fields usually feel like, with all these colour variations. I aim to make this more obvious later, with smaller chunks, but I have to keep working so it is going to stay as it is right now. The colours were taken from diverse screenshots from the movies, and then altered a bit. It is true that they are brighter than other Studio Ghibli movies, but I am happy so far with them.

In the last weeks the project has grown a lot, and as usual, I have left the documentation behind. Later I got Covid, so it has been a couple of weird weeks. So many changes have happened to this project, I have uploaded a video to YouTube where all this can be visualized:

Field video

The images are quite low resolution, so they can be found here.

I am going to list all the new improvements in a small list, so we can talk further about them later. Even though I am late with this, I am not overwhelmingly late, so without further ado.

Colour rework

Clumps of flowers

New ground texture

Trees

Grass Wind

This has been fun, although Unreal has given me so many problems that sometimes were fixed just by restarting the engine. The base of it comes from this YouTube tutorial.

The main feature of this project is the huge field, and therefore there must be other things on it apart of just vegetation. But there is a problem, when placing them the grass collides through them, and leaves an ugly look.

To avoid this, we make use again of RVT to send information, in this case through the Roughness channel, when the object is located.

This way, and by setting a default value in the landscape material, we can know where there is an object and just offset it downwards until it is not shown.

So as it can be seen, we now have an object that occludes the grass.

Recently I have been working on vegetation for the project. One of the main concerns that I had was the look of the grass and the ground. You can usually see the seem between both. To manage this I made use of Runtime Virtual Textures, which send the information of the ground texture towards the grass on itself.

Thanks to this I could use a texture floor to add variation to the grass.

That variation really sells that this isn’t a flat environment. The main problem I have right now is that I have set every blade to be unlit, as it looks better, and therefore making shadows on them is going to prove difficult.