CG General: a diagnostic add-on that finds what's wrong with your scene before you render it

[MartinZ (Martynas Žiemys)]

@TheFridge

But.. doesnt mean the good lighting is what you wanted. It might be technically correct but not creatively.

There is no ‘correct’ in art anyway. Its just a perspective.

I don’t think this is what was meant. These are the hardest concepts for AI fans. Image is about communicating something visually. Lighting doesn’t have rules or constants but it does have meaning. For very simple example, you communicate “warm” with yellow/orange/reddish lighting, right? It’s like words in telling a story. If you let AI control stuff it’s not your story any more. So yeah

@sozap

But it’s possible to analyse how creative choices serve the purpose of an image, how they convey the story you want to tell, how they relate to reality. These for the most part could be objective.

This is about visual language. AI talks nonsense in this language. Even if it was good at this, this doesn’t matter, because it does not know the story you want to tell, so it’s not your story any more if you just ask AI to “fix the lighting”. If you don’t communicate what you want to say and provide no context, it makes no sense. If you do communicate what you want to say to the AI, it works way better. Amazing, right? Not really, then you just have an assistant with cognitive abilities of a toddler, so that’s not helpful either.

[joseph]

@TheFridge

The thing is that lighting is actually quite mathematical and there exists a whole science behind lighting theory. Its the mood or emotion that the artist is trying to invoke which is what is subjective. There is good and bad lighting - objectively. But.. doesnt mean the good lighting is what you wanted. It might be technically correct but not creatively.

There is no ‘correct’ in art anyway. Its just a perspective.

But this would help to atleast get the technical part right. You dont have to apply all its suggestions, nor are you stuck with what is applies. Creative direction is always gonna be in the users control , and thats what matters. But i think its nice to see a slightly diff take on lighting at times.

What about your tangents and leading lines going nowhere is “mathematically” or “technically” correct?

[TheFridge (Muhammad Haris Saleem)]

That’s just me. Lol. I made the scene. And I’m human. Still learning every day. So if I don’t have perfect composition or something that should be okay. I didn’t even think about the leading lines at the time.

But I see what you mean with the blue and orange at tangents. Could be improved upon…

[TheFridge (Muhammad Haris Saleem)]

I get it. You just hate AI. I can see that from your comments. The post was not to invite roasting and hate. If anyone has constructive critisicm it’s welcome.

Otherwise it’s just an AI hate thread. You can take it or leave it. No one is forcing you to like AI.

[MartinZ (Martynas Žiemys)]

@TheFridge

I get it. You just hate AI.

I don’t like it, you are right. Not necessarily the technology itself, but mostly the way it’s used. It’s amazing technology, but I think a lot of people use it for purposes where it fails the most. This is one of those in my opinion. And I think there are a lot of hidden dangers in relying on it too much. It may seem like it’s a good idea to have a “second pair of eyes” with it, but I think there are hidden dangers in trusting it. It fails to see and consider the context and results are often worse than one might expect and very often that’s worst in situations where one simply lacks knowledge in the subject - so new users for example, which is sort of the target audience of this. I think it’s important to talk about it, because it may be misleading. People may spend time and effort and money believing it’s an amazing tool to only later find out that it has serious issues. I think it’s very important to talk about those issues.

No matter my feelings about AI, I would think my logic arguments still stand. Some things I discussed have nothing to do with AI. Like for example the perspective that you can just make lighting better without considering context, purpose of the image and visual communication. That has little to do with AI if you think about it. It’s more than pure hatred or dislike. Those are logical arguments.

[BlenderTimer (Daniel Roberts)]

I personally think this is a very interesting addon and will likely be very useful for a fair number of people. I’m actually a little surprised it’s getting so much hate. LOL

Also, I don’t mean to sound rude here, but a lot of comments from people in this thread sound like they have very little knowledge and/or experience with AI. It may be a good idea for them to try out various AI systems and learn how to use them well before deciding AI isn’t good or doesn’t make sense for this use case.
AI can be extremely powerful for quick problem detection, such as lighting issues. In fact several months ago I put one of my images into AI and it found a fixable issue which none of the dozens of other people that viewed it even mentioned. And usually, asking AI is actually more reliable than asking people. Sure, if you have access to a professional art director at your “beck-and-call” that’s likely going to be more reliable. But most of us don’t have many people around us, and those we do have around us are often the same kind of people who will make a PowerPoint with a JPEG-compressed paper background, 500 different fonts, and psychedelic colors. That’s not really who you want telling you how you should make your art better.
And, importantly, the implementation here of AI recommendations rather than AI outright making the art how it wants it is arguably the best method of perfecting art. You make the artistic decisions and decide what to do, but the AI gives you good ideas which you can try, and you can rarely have too many good ideas.

It’s the same reason we have things like low-fuel warning lights. It’s not that mechanics aren’t needed or useful, it’s that not everything needs to be done by a professional, not everything needs to be done perfectly. If you know a problem exists, it’s up to you if and how you want to fix it.

[sozap]

Well ok thanks a lot I’ll definitely check those checks it’s interesting ! I’m lacking some bits of brain space now to fully study that… but definitely later !

As for the scene analyse to deduct a lighting, it’s also quite interesting indeed, can you outline roughly how the algorithm work ? How it considers a good lighting ? More examples could be interesting to get a feel of what result you would expect given some input.

In your main example, both before and after artistic intentions are quite different, it’s weird that it completely removed most of the first proposal…
We can consider that this is a bad lighting because it’s front light, but this could work really well for a flash photography and this could be a starting point for a very interesting image, in real case scenario I don’t think having such a drastic difference would fit.
Maybe something like you have 60/90% of the stuff figured out but it helps with the remaining 40/10% would be more interesting as a real life example…

Indeed I wouldn’t send my scenes. Maybe trying some of the lighting challenges as experiments :

3dRender.com: Lighting Challenges

Would be fun to see how your algorithm manage with various test cases.

@TheFridge

I’m always glad to receive constructive feedback or if you have any ideas that could improve on this. Something people find more professional or useful?

Hum… As said, sanity check is part of a professional studio tool, generally checks are added as problems occurs. Generally they have two modes : 1 diagnose 2 fix when appropriate…

As for the lighting tool, well if you’re a professional doing lighting, you know how to light right ?
So while the check tool could fit an intermediate or advanced user, the second one is likely to fall really into the beginner category.
It’s also worth mentioning what kind of industry it’s targeted at… maybe that can work for basic product viz where it’s recurring lighting pattern but in this case, probably presets would fit better as you can test different stuff and then tweak the result.

For basic arch-viz maybe that would produce something… Anyway what am I seeing in your test example, is everything adjusted automatically by AI ? Or it gave you guidelines you tried to follow ?

For illustrative work, well then it should understand creative decisions, context, style, there are also a lot of constraints that probably the AI can’t integrate in it’s diagnostic.

In the end how good is your tool, or how relevant it is to a professional depends on how clever you tool is and how meaningful the feedback is. It can be fun to run that last example image through your AI as see what it has to say, then I’ll give you some feedback on that !

You also mentioned iterations and stuff like that but it’s to me a mix-match of various feedback you can get. If you work for an art director, they have a vision and they help you to get there, and yes there are some iterations and back and forth.

When you seek advice from peers, they give some feedback but it’s our job to understand and process that. In that case they point out a problem, but you have to understand why it’s a problem to them and what is the most appropriate way to solve it. From there iterations append but you need to provide something that seems to you an appropriate solution to that issue.

All in all, while AI can find some stuff, they have limited understanding of the subject, the purpose, and also the history of the project and obviously they don’t react emotionally, which is the point of our work. So yeah you can see what it has to say but it cannot act as an art director or even peers. I don’t think you can learn a lot by having an AI as a teacher or supervisor. At least it goes down to how clever it is. So maybe getting some understanding of it’s logic might help answering that. From there it’s easier to see who is going to benefit that and at which point our skills starts to outperform those of the AI …