Hacker News Reader

Comment by armchairhacker | Hacker News Reader

How much of these assets did he make himself?

Regardless this is extremely impressive. There's nothing to distinguish this from something made by a group of professionals with infinite time and resources.

View on HN

They likely created a large portion of these assets themselves. There might be some kitbashing but the amount of detail here is certainly doable for a motivated 16 yr old.

Theres a lot of creative ways to add details using scans, texture projection and scattering.

Or, you know, coding.

From what I understand, Blender's nodes increasingly replace the need for code with each version.

Yeah, geometry nodes almost feels like a cheat.

On the other hand, it's like Scratch and from my distorted worldview exposing it as code would be more straightforward.

My brain just seems to prefer having a tool to reach for with a clear purpose. For me, availability of a thing that does what I want is generally the difference between being able to do it or not. I've tried learning to code on my own, but I always hit a "now what?" situation since the available tools for the things I make are already better than I have any interest in improving on. I'm not a programmer by interest and don't foresee ever being one by trade.

I wrote some words on the topic: https://kyefox.com/2022/08/05/learn-to-code-or-dont/

Node editors like Blender's are the best thing for me: you get all the tools, but they're composable with other tools. Your hammer can be a screwdriver. Worst case, I need a new tool no one's figured out how to make with nodes, but odds are someone already wrote that code as a plugin. Looping back to the "now what" issue: it already exists in a state beyond anything I could get up the motivation to improve on.

Well creating a node and hooking it up to other nods is most probabily something like this under the hood:

  pd_node = PointDistributeNode(type = PDType.RANDON, seed=149)
  scaleRandomizerNode.geometry_input.connect(pdNode.geometry_output)

So far it's perfectly represented by node editor.

How about creating a dynamic number of point distribution nodes and adding them all together? Trivial in code, most likely impossible in node editor.

The most flexible UI is a good programming language, anything else comes with flexibility compromises in favour of something else.

Thanks for this comment, I find that a really interesting perspective. I'm the exact opposite -- I'm comfortable with code, and always find node editors really awkward to use (but I definitely feel I'm missing something because they're undeniably super productive in many situations).

They're super productive when you're unfamiliar with the subject matter, do it once only or can't code at all.

Anyone can watch a couple of videos, launch Blender, start with a cube and work it into a chair in UI. Coding that for just one chair will take almost anyone more time.

Now if you need to do this to make 10 slightly different chairs where changes need to be made early on in the process, code will win even with first timers.

with stuff like 3d modelling code you're kinda skipping straight to some of the more hardcore coding. 3d transformations and algorithms are no joke, thats university and master's degree level maths. Don't feel bad if you don't get it, a huge number of professional programmers wouldn't either. I write a lot of software myself close to the metal as a robotics engineer and ive tinkered with blender code but id definately need a lot of practise to feel comfortable doing that without glueing together tools

Totally. 3D geometry is hard and when you're coding there are a bunch of different ways a minor error can mean absolutely nothing appears on the screen. Just learning about those different failure modes takes a while.

The other thing I find (but maybe this is the same with node-based approaches too!) is that there are a bunch of arbitrary coin-flip choices you have to make when doing 3D (e.g. left-handed versus right-handed coordinates). At a fundamental level it doesn't matter which choice you make, but your decisions have to be consistent, otherwise things can get flipped around, or often just vanish entirely. The sneakiest problem is when you unknowingly make two wrong choices, and they mostly cancel out except for some weird edge cases that happen to exercise the exposed code path.

Good point, i like the balance houdini has here where you can drop into vex when you want.

Hard surface modeling in blender is pretty easy if you go the kit bash route. You still have to have some concept of what you're building. You also still need environments, rigging, lighting, camera work, compositing, and post.

I also wonder about the sound design which is very well done, but seems to be a very different field than 3D?

Reply on news.ycombinator.com