Storage space requirements for sprites are incredibly small.

A 24x32 pixel PNG can be as small as(or smaller depending on what the image is) 368 bytes.
Now that is just as a PNG, you can have custom formats that could be much smaller than that.
A FLIF file of the same image (just a quick conversion, I haven't played around with FLIF as much as PNG) is 215 bytes.
Both files are lossless (FLIF can do lossy too).

Just some cool information I wanted to share.

Huh, I've never heard of FLIF. I'm familiar enough with PNG though. Anyway, there are 2D games with higher resolution sprites as well (e.g. Cuphead), which is why I didn't say 2D games always use less memory. I reckon storage space requirements ever for large sprites with large areas with fairly uniform colours are much smaller than more realistic and varying sprites though. Suppose someone made a fairly realistic-looking 2D game with good animations, however, and I would expect it to use a fair amount of memory. Please do correct me if I'm wrong, of course!

FLIF is pretty cool from what I have seen and tested.
It is FOSS too, so it gets extra points in my books.
http://flif.info/
The main issue with using FLIF: it is almost universally unsupported.
Hopefully that changes someday; it really is cool.

I tested a few sprite sheets from Cuphead:
https://www.spriters-resource.com/pc_computer/cupheaddontdealwiththedevil/sheet/98215/
https://www.spriters-resource.com/pc_computer/cupheaddontdealwiththedevil/sheet/98516/
https://www.spriters-resource.com/pc_computer/cupheaddontdealwiththedevil/sheet/97147/

Psycarrot
Download: 7.13MB
optipng: 5.84MB
FLIF: 3.73MB

Moe Tato
Download: 3.85MB
optipng: 3.10MB
FLIF: 2.13MB

Botanic Panic
Download: 1.74MB
optipng: 1.47MB
FLIF: 1.25MB

I am pretty sure Cuphead's download size is about 2GB.

With sprites you can make them modular; Terraria seems to do this for characters and stuff.
Sprites have a pretty big downfall being: it is hard to compress them in a lossy manor.

3D games have the huge upside of being able to use interpolation to figure out what goes between animation key-frames.
3D games (and some 2d ones) can also use procedural animation techniques (rag-doll, IK systems, ect), further reducing the need for additional animation data.
A downside of 3D graphics, is easily, the fact that you have to store all of the textures for each object in a scene.
With a 2D game you only need to store the data for what can be seen from a fixed perspective (generally speaking).
Generally speaking players are not moving a camera around in a 2D title, so you can save a lot of storage space, and artist hours by not authoring the unseeable.

Personally I am very excited for the future; there are some very cool application of neural networks, I really want to see incorporated into games and game development.

Imagine the possibility of a neural network powered: sprite animation interpolation system.
Where the developer only packs in a set number of key frames, and lets the game engine generate the missing frames using a neural network.
Or.
A neural network based upscaler (lookup waifu2x) built into the game engine; allowing the artist to pack lower res assets into the download, to let the engine unpack the final assets when required.
Or we could get some really neat texture compression system that uses a neural network to further enhance compression ratios, somehow.

Volta and Navi really help me keep hope in the future of video games.

I can't speak from personal experience. I don't do any coding. One of my best friends does though. He tells me UE4 is much better to work with than Unity.

Honestly, it depends on what you need. Neither one is "the best" nor does neither one "perform better" in all situations. Unity has an advantage in mobile, while unreal has an advantage in AAA.

Unity is quick for programmers to pick up and get things running. Unreal has a learning curve of sorts, but with C++ scripting instead of Unreal script now, it is much easier than before.

Unreal has far better tools, especially for art, animation, world building, visual scripting, etc (right out the box). Unity is lacking in that area. If I were to guess why that is, it would be because Unreal was created by a studio that makes AAA games, and as such, they continue to use it and make tools to create said games. Engine creation, not game creation, was (and is) the main goal of Unity's creators.

Unreal has the awesome visual scripting system (blueprints) and how it ties into the c++ side is great as well.

Unity has a much better shop for plugin's and addons.

Unreal has the better renderer, no competition there.

Unreal provides the source code by default now, which is a great feature for programmers.

I picked unreal in the poll, because I personally like it the best, and like working with it, but where I work we are using Unity for our project. Nothing wrong with that, it is a great engine as well. It all comes down to the needs of the project, and personal preference.

A 3D model would be unnessecary more complicated to pull off in what they wanted to achieve. They would require a mesh of polygons transforming to the object they need to go along with the art style. 2D was the way to go.

I'm not a dev, so take this comment for the absolutely nothing that it is worth...

Both Unreal and Unity have various strengths and weaknesses, as do all other engines. So, it's not a case of which is better. It's a case of which is better for a given purpose - in this case, a given game. We knows this because developers say it all the time.

Cool. I looked up some info about the format, and people were worried about patents and encryption and decryption speeds. Encryption shouldn't be a huge issue so it shouldn't matter if it's slow, and it doesn't sound like decryption is too slow either. The main problem with patents seems to be uncertainty: The team hasn't done much, if any, research on the situation, beyond knowing their starting point. I hope they clear up the patent situation because the format sounds awesome.

Neural networks are truly magical. It's as if they can do anything better than existing solutions. I know it's not true, but they're very versatile, powerful, and efficient.

They are pretty wonderful.
Here is a sprite taken from a Cuphead sprite sheet, upscaled by waifu2x (a neural net upscaler).

Untouched image:

https://image.ibb.co/kKCQFR/Psycarrot.png


Scaled with waifu2x:

https://image.ibb.co/cFdMpm/Psycarrot_waifu2x.png

Scaled and denoised by waifu2x:

https://image.ibb.co/kiwVh6/Psycarrot_waifu2x_denoise.png

That's a managed language (C#) for you. Unpredictable performance comes with that kind of language (mostly garbage collection related). Allocate and manage memory yourself, game programmers. Understand memory lifetime and memory management is dead simple, and usually more performant than GCs.

Mega-Textures for 3D games take the cake. A reasonably detailed compressed (DXT; 4 bits per pixel) texture (128K x 128K) is like 2 GB, and 16 GB uncompressed (RGBA8, 32 bits per pixel).