Hacker News Reader

Comment by hu3 | Hacker News Reader

Looks good but rather too foggy and not as crisp as the original which is much more detailed. Video artifacts notwithstanding.

blender: https://i.imgur.com/0BjMWi3.jpeg

original: https://i.imgur.com/GGwVibG.jpeg

View on HN

IMHO it's not even close.

Agreed. As a tech artist, I'm 100% positive I could get a bang-on reproduction using Arnold or some other really solid photorealistic rendering engine, but it would probably take longer than using real cameras, practical props, and doing a bit of massaging in post. The modeling, color, camera angle/focus length/etc. are all really easy. Tweaking the subtleties in the light, glass shaders, fog, etc to get them good would take quite some time. It would only make sense if it needed to be animated or you needed a bunch of versions with modifications. Definitely a 'use the right tool for the job' kind of thing.

TBF, the blender version you reference was from an animation, not the first single still render.

blender (frame 0): https://i.imgur.com/XIV8Hma.png

original: https://i.imgur.com/GGwVibG.jpeg

Also, it includes a caveat:

> Unfortunately, due to the nature of volumetric rendering, I was unable to economically render the final animation at a high resolution with enough samples. So I had to resort to denoising, which sadly degraded the image quality and made it a bit flickery.

> Perhaps, with enough computing power, I'll be able to return to this project in the future and provide a cleaner final render.

And so the remaining 20% of the work will take another 80% of effort.

Classic

No, not really. Increasing the volumetric samples + bounces size is fairly simple. It's just a value in the rendering tab.

Reply on news.ycombinator.com