1. Research
1.1 Tools
1.2 Other artists and companies
2. What I found interesting
3. Conclusion
1. Research
I chose this goal as my Applied research topic as well because I would have to read up a lot on this topic already. Trough work and media I can see other people start to make use of AI technology to improve their work. What I rarely see artists using AI in a positive way. Most of the discussion is on copyright laws and the negative impact but I wasn't to find a useful and mindful way to use AI for my own benefit as a 3D artist.
1.1 Tools
First, I started by listing the processes in my workflow that are the most repetitive and annoying:
background editing and photobashing
It is a useful skill to have, and some artists solely rely on photobashing for their artstyles. To me it's a useful tool to add texture to an environment concept but as a character artist I often don't want to spend time on backgrounds but at the same time I still need a good background to support my concept.
A popular tool that was recently made available is Photoshop’s AI tool. It is still in beta, but I could try it for free.
It is definitely not perfect, but it solves some very time-consuming problems. Mainly, if I want to use an image for my background it almost never fit exactly my needs in colour lighting and composition and the aren't that many free image galleries if I need to use it for commercial purposes. With the AI cut tool I can completely change the image quickly.
It is good at making big changes that alter the image, but small details don't look particularly good.
It gets a bit confused and changes the colours and lighting too much.
Sometimes it can't fully understand the prompt if it is extremely specific.
3D model generation
I prefer doing my ideation in 2D, but I can't do AI research without trying out OpenAI's tools. I read the documentation on Dall-E and how it works as well as some discussions online on their use of other artists' work. I don't think the tool is good enough or time saving enough for to be used yet and the copyright issues make it even more unlikely for my use.
I investigated it purely for research purposes. The models are not good for only use at their current state at least not in the current state. I considered using them for ideation but that also doesn't give good enough results and it doesn't seem to be speeding up the process at all. I found research done by a group of anonymous artists that reached the same conclusion.
I would consider using the tool in the future if the models it generates are more workable (have better cleaner topology). I can see it creating small hard surface parts like rings, bolts, or robotic parts that I often look for online to save time.
Retopology
Retopology is by far the most repetitive and annoying part of character creation. I think an AI tool that can perfectly fix the topology of a character and make it ready for animation would be the most useful tool for any 3D character artist.
Sadly, there isn't such a tool developed yet because AI technology is new. According to programmers teaching a neural network to do good retopology is harder than teaching it other things in other sectors that need equal attention.
There seem to be some developers already trying to experiment and test their tools, but it is something to look forward to in the future.
What is available right now are automatic retopology add-ons in different 3D programs. They are not AI based but it still seemed like an interesting thing to explore. The best one to try according to artists online is the zRemesher tool in zBrush. Luckily, I have zBrush on my laptop, so I did a test.
I was not too happy with the results. It seems like it is saving time, but it really depends on the surface and how I separate it. So, if I consider how complex my models usually are the preparation needed for the zRemesher to work adds a lot of extra time to the workflow. And even then, there are some mistakes I manually must fix later. It is an interesting alternative to consider but I don't think it makes the work much easier than simply doing the retopology myself by hand.
At the current state of automated retopology I wouldn't use it.
UV unwrapping
I personally don't have a problem with is I find the process relaxing, but I would like to have automatic UV unwrapping when I'm short on time. Sadly, the same as retopology – the tool does not exist yet.
Texturing
There were 2 interesting things I found about texturing a 3D surface.
5.1 First I was surprised to find an online tool that can generate textures directly on your model using a prompt.
I will use the tool if I need to make small details to an environment that don't need to look too good and are only giving an accent or silhouette to the overall scene. For character creation it does not seem so useful because mistakes and patches on a character are much easier to spot. The quality of the textures is low, and it is clearly meant for small fast details.
With that in mind I can see myself using it in some cases when I really don't have time to spend. For example, sometimes there is a small piece of armour like a belt or Sci-fi part that needs some detail and id usually either must draw it in photoshop or for a more realistic look I will have to look for a copyright free image online. Both options are tedious, and the tool would be useful in that case.
5.2 The other interesting use for Ai in the texturing process I found was creating tillable and complex texture maps for procedural textures. Often when creating a base for a procedural texture or a mask even in Substance painter I need to use the presets. If I need something specific there are few free options for images. In that case I can use Dall-E to create tillable base maps for example. In addition, I can use Ai to upscale the resolution of maps as well which is extremely powerful.
2. What I found interesting
LOD or Level of Detail is another part of creating a good character for any game. This is not necessarily something I struggle with because there are automatic LOD generators in most engines, but I found an interesting new tool developed by NVIDIA.
On the video they show many more uses for the technology but from my perspective as a character artist it would be useful for having beautiful objects at any level of detail while saving performance.
Animation is also not exactly part of what I do as an artist, but I finds myself having to do it more regardless of my specialty. 3D animation is a very tedious process that needs a tool to speed it up. Especially in cases like motion capture where you would expect to get clean animation data but there are always problems even with the mostly expensive motion capture setups.
What I was interested in was a tool that can help with finding and separating animations into keyframes, so it is easier to edit them but what I found instead were 2 different more interesting approaches. In the case of animation, the most progress is being made in translating video into animation data. There are several tools that can do that, and they look promising as they already gave reliable results in the stage they are. They are also much more affordable than the more tradition ways of motion capture.
The other interesting tool is Cascadeur’s automated physics. The idea is that but creating keyframes and moving them around the software can intelligently connect them into an animation that takes physics into account.
I had the chance to play around with the tool and it is useful for big movements like dancing, jumping, acrobatics, running that require physics to look realistic but smaller movements like an arm swing still must be animated by hand. This is because the way that the software works is my calculating the canter of mass of the whole character and compensating by adding the proper leg movements.
3. Conclusion
I spend time on this personal goal throughout the whole semester and I think it was time well spend even though it was not needed for our IMT&S project. I think all the things I've learned on the topic are going to be useful to me personally for the portfolio pieces I am already making on my own.
Out of this research I would implement in my workflow Photoshop's background editing tool because it is a fast and effortless way to elevate my final render when I want to present my concepts and especially I when I want to post them on my portfolio page.
The other tool I find powerful is using image generation for masks and texture maps in substance. Simply generating a map instead of using the presets will already elevate my style and texturing skills.
And finally, I would keep using Cascadeur because of all the animation time that it saves. It can't do everything but as a junior animator or saves me so much time and makes the end results much better than simply animation in Maya.
Comentarios