Artist Ostap Gordon just released a Russian yard in the winter in Unreal Engine 5.1 with high-quality assets created with Megascans and Blender, fully dynamic lighting with Lumen and Nanite for all objects. Check out this ultra realistic walk recorded with a RTX 4090 in a 4K monitor.
The lighting is good. I think there’s too much ambient light to feel real. The trees are great. Very cool to see the detailed trees far away. I would like to see more extreme versions of high quality detailed nanite scenes.
It does have this effect on me of every now and then it feels like it’s blending into a real life camera shot like you might expect a game to transition from pre rendered cutscene graphics to game ui.
As a Minnesotan I can tell you that even in light snow, looking up at a street light at night there is a very very noticeable effect where the cone of light from each street lamp looks like a miniature blizzard. In the dark you cannot tell whether it is snowing or not, but within the light of a street light even minimal flurries look like a snow globe.
I did not see that effect here and it really broke the immersion for me. Every time they looked up at a street light I kept expecting the "snow globe" effect to pop in, but it did not.
Further the snow on objects is too smooth and flat and uniform. It may melt and refreeze into that kind of shape after a few days, but in this demo it is actively snowing. The snow on objects would have a pronounced fluffy look like a sweater.
This is fantastic but it also clearly points on how deep a rabbit hole it is to try and do 100% simulation - a fractal challenge? There are so many tiny details that give it away, and I realize how technically complex it can be to solve each one of them.
Such as the wind not matchig with the lack of tree movement, the way the camera moves, the repetitiveness of the walking sound, how the wind sound doesn't reflect how it would be blocked and dampened by buildings and snow, how snow doesn't pile up, how the tires look blocky, etc. (and the final challenge, hold out your hand and look at the snow flakes ;) )
Add lack of IEC lighting, how it's exposed/graded in a way that makes it easy to spatially orient - which is useful as a player in a game, but not realistic.
Your wind and camera movement comment in particular provoked a feeling of curiosity. We should be able to statistically model both in a way that their movement in game is optimized toward minimizing the KL-divergence between empirical and simulated motion.
It looks nice, but it's hard to relate "ultra realistic" when all reused assets have identical snow accumulation patterns on them. Every tire in the ground looks exactly the same. Every car of a type looks exactly the same. And so on. There's no natural dynamism whatsoever.
11 comments
[ 5.2 ms ] story [ 61.2 ms ] threadArtist Ostap Gordon just released a Russian yard in the winter in Unreal Engine 5.1 with high-quality assets created with Megascans and Blender, fully dynamic lighting with Lumen and Nanite for all objects. Check out this ultra realistic walk recorded with a RTX 4090 in a 4K monitor.
It does have this effect on me of every now and then it feels like it’s blending into a real life camera shot like you might expect a game to transition from pre rendered cutscene graphics to game ui.
I did not see that effect here and it really broke the immersion for me. Every time they looked up at a street light I kept expecting the "snow globe" effect to pop in, but it did not.
Further the snow on objects is too smooth and flat and uniform. It may melt and refreeze into that kind of shape after a few days, but in this demo it is actively snowing. The snow on objects would have a pronounced fluffy look like a sweater.
Such as the wind not matchig with the lack of tree movement, the way the camera moves, the repetitiveness of the walking sound, how the wind sound doesn't reflect how it would be blocked and dampened by buildings and snow, how snow doesn't pile up, how the tires look blocky, etc. (and the final challenge, hold out your hand and look at the snow flakes ;) )
Your wind and camera movement comment in particular provoked a feeling of curiosity. We should be able to statistically model both in a way that their movement in game is optimized toward minimizing the KL-divergence between empirical and simulated motion.