Taiga & Giant Spruce Biomes Completed

Hey!

Go check out the new Youtube video: https://youtu.be/CNZdTkYl9O0There's also a second video for next week, but you guys can watch it right now: https://youtu.be/Ie61IwXUQgI

The map from the videos is a normal randomly generated world - Not custom built! Walking through the forests and exploring these biomes was a lot of fun because there are so many different areas with varying density/types of vegetation. This also results in very interesting lighting/shadows. Would love to see what areas you can find because almost every corner of the forests looks different.

Shaders:

Almost all the shots from the videos were taken with a modified version of SEUS Renewed 1.0.0. (changes: waving vegetation, reduced skylight, fixed POM issues).

Downloads:

Renewed E2: https://www.patreon.com/posts/seus-renewed-e2-20912484

Renewed E1: https://www.patreon.com/posts/seus-renewed-e1-18917158

Renewed 1.0.1: https://sonicether.com/shaders/download/renewed-v1-0-1/

The older versions have nicer fog/mist in the forest but you need to increase the sun intensity and image sharpness a bit. (and maybe compensate by lowering exposure to get a better skylight vs sunlight balance)

The shots with the fog/mist lightshafts were taken with the shader ComplementaryUnbound v5.5.1. Make sure to change the settings for the lightshafts because they are usually not always there.

Videos:

The camera movement in the past was way too "cinematic". Sure, that's suitable for the Rendering texture stuff but for the gameplay textures it feels more natural to have a first-person view with real mouse/keyboard movement.

The videos were rendered with the replay mod to ensure 60fps at high resolutions. But it seems there are some issues because the camera/player movement feels a bit choppy. Not sure if there's a way to fix that.

Also: The videos are still recorded on version 1.14.4. because newer versions seem to have some shading issues for grass and there doesn't seem to be a way to fix that. Going forward it seems that a good strategy would be to upload way more videos to get some attention. e.g. multiple videos for each topic/theme spread with some time inbetween. Maybe one long-form video every week and a few YT shorts inbetween.

Textures:

The spruce stuff and podzol is old test stuff from 2020/21 that was never really released publicly. The only new stuff: fern, large_fern, red_mushroom, brown_mushroom.

Spruce log: Consists of 3 textures (ctm repeat) stitched together from real photos. The logs require custom models in order to get the ctm working correctly when the logs are placed horizontally. The normal map details are extracted from the luminance of the base texture and the "flat" parts of the bark were painted manually. Turned out pretty well. The height map is also done in a similar way.

Spruce leaves: The idea originally was to make the branches connect to each other in a more natural way. But that's tricky because spruce leaves are not as uniformly packed as in oak/birch. So to make the branches span a larger distance, they need to be in random orientations without a fixed "center". But making the whole block too big also causes issues (e.g. ptgi voxelization), shading glitches, and the block becomes difficult to interact with (place/destroy). So it's a weird balancing situation as always. But looks okay with Seus Renewed 1.0.

Podzol: Basically a placeholder texture. No PBR. Bad tiling. Just a picture slapped on a block with some random rotation. But it's nice because it makes the whole giant spruce biome look very cozy due to the orange warm light it gives off when hit by the sun (if your shader supports GI). It's a texture that would also work in 64x without anyone being able to tell a difference when you walk through the forest randomly.

Fern: It doesn't use biome tint because that usually looks ugly. The chosen hue and saturation seem to work well across all biomes. There is an alternative version of the fern in the R17 pack (in high resolution) with slightly different colors that are more suitable for rendering-related builds.

Why the lower texture resolution for the gameplay stuff?

High pixel count doesn't make a texture look good or realistic. Patrix showed with his pack that for gameplay, large-scale variation is far more important than making individual blocks realistic. 256x is enough for regular gameplay. 128x is also fine if you want more FPS. Considering this, the public/free 128x version offers a lot of value and is perfectly fine for most people. The same was also true for the 512x "rendering textures." Most people wouldn't even be able to tell the difference if the videos showcased the 512x versions instead of 2048x.

Texture downscaling considerations:

Texture upscaling is difficult for obvious reasons: the algorithm has to invent new details. But texture downscaling is also quite tricky and there is no solution without trade-offs. In the past, the "rendering textures" were downscaled manually by using a mix of techniques and then adjusting sharpness, contrast, etc., to match the original texture as well as possible in-game. But that's an insane amount of work because every material/texture needs different treatment, and doing this correctly for normal maps would be even more difficult.

Patrix would probably never do that for his pack due to the insane amount of textures. It seems like he uses regular nearest-neighbor sampling to keep it simple.

Nearest neighbor sampling: The algorithm just picks one sub-pixel and reuses the exact same data. This is perfect for maintaining the alpha channel and to make sure that the encoded normal map vectors are still correct and normalized. The issue: if you downscale 2048x to 64x (as an example), you end up with a super noisy, unusable image. That's because the algorithm often picks a pixel that is completely different than the surrounding pixels. In the downscaled version, this pixel's color then covers a huge area that does not represent the corresponding area in the original texture.

An easy solution for this is to just sample multiple sub-pixels of the area and average their color/values. That's basically what algorithms like bilinear sampling do. This indirectly maintains a lot of information from the original texture but ends up looking very soft or blurry. To fix that, you can just apply sharpening filters to match the perceived level of detail/sharpness of the original. That worked well when downscaling the 2048x to 512x. But downscaling to 128x or lower is way too ugly.

There are other, more complex algorithms that aim to maintain the perceptual sharpness, but in the "gameplay texture" cases, it did not work well and looked similar to just bilinear + sharpening.

Conclusion: Good old nearest-neighbor sampling is the way to go because it's easy to automate and you don't have to worry about anything. You just have to accept a little bit of noise. (But it's not that bad when going from 512x to 128x.)

In the end, the best approach would be to just re-paint the texture similar to pixel-art artists. But obviously nobody wants to do that. Hopefully AI will offer a good solution soon.

The Downloads can be found on the Download Post:

https://www.patreon.com/posts/public-downloads-127596976