Clients and AI

[louisryko]

This is more of a rant than anything else, but I'm also looking for some community input regarding an on-going issue I'm dealing with...

I have a client for whom I create a lot of residential 3d imagery for. This client, let's call him Jay is extremely picky with detail and precision. About 5x more than any other client. Every single pixel and mm in my scenes has to be perfect for them. Like, the tiniest bit of flicker in an animation has to be fixed, or a coffee cup in the back of a living room isn't rotated 5degrees more 'correctly'. Or the a blanket isn't fluffy enough. You know the deal...

Jay is also constantly pushing me to investigate AI. Like, really pushing me. Every conversation includes "ok, and can we somehow make that job faster with AI?"

Of course, I use AI touch-ups, particularly on scene people. But I still mostly get Max + Corona to do the grunt work.

What Jay fails to understand is that AI is not a one-click solution. It's not a band-aid. It currently doesn't do the things he expects it to do. And it certainly won't reach the level of detail and precision he expects in my regular cgi stills + animations. At 7k resolution, too!

Now, in my latest conversation with Jay - where we're discussing modelling 200+ unique apartments for glb export (for web-viewing) he insists I do thorough research to ensure that some AI software isn't out there that will save me from manually importing each CAD floorplan and modelling each apartment. He believes we can get AI to not only model each apartment (accurately, and consistently, mind you!) but also with lighting, furniture and all textures and make it suitable for low-poly glb export!

I see Jay 'like' and 'share' so much 'AI developed' stuff on LinkedIn. Half of which is fabricated (clearly not AI to my cgi-trained eyes) and the other half is just such poor quality. This is one of the roots of the problem, I think.

Of course, this is all coming in the name of time and cost saving. Which I understand. But there are doing things right (the way the client expects the final result to look like) and doing things quickly with AI.

Ugh. How can I convey to Jay a more accurate reality here?

Or am I simply wrong? Is all this possible in AI these days, in a time-saving, accurate and reliable manner?

I'd love to hear other experiences if there's any out there. Cheers. /rant

(Reposting from a throwaway account - I hope that's OK?)

[TomG]

You are correct in your understanding of AI. It is fine for generalities, but it fails when it comes to specificity. And it sound like specificity is a key requirement of your client (as it is for many).

If you want to "edit" an AI result, you just have to roll the dice again - you can't make specific changes, like "rotate the cup an extra 5 degrees" is precisely what AI can't do. Or if you had AI fill in a high res version to replace a low rez model (e.g. people), and you want the person to be identical but change that jersey for a jacket... no can do, you will get a different person where features etc. change as well.

AI has its uses for ideation, and then for "refining things where specific details don't matter", but that's it for the present time. Also I'd say this is pretty fundamental to how AI works, so "present time" could mean until there is a radical and as yet unseen shift on how things are done.

This is why Chaos has very specific uses for AI and we are not just throwing it into everything everywhere :) We want to use it where it is genuinely useful, given the actual jobs that our customers do ;)

As to how you convince your client, well that is harder. You may want to search for articles on the problems AI has with specificity, find some videos, tutorials, and even papers that describe that shortcoming.

[TomG]

PS and your case is not helped by all the hype that comes with AI, both from the companies that make it or use it in their products or offerings (as many more companies include AI but do not make it in any way), but also from users who have a blinkered view - maybe for their clients and projects, AI is just fine; or maybe they've invested too much in it emotionally or financially to afford to have any doubts or admit any shortcomings; either way, you get a lot of vocal "rendering is dead!" stuff flying about (just as "GPU rendering has killed CPU rendering!" has been said for oooh the last 15 years or so....)

[louisryko]

Thanks for the reply. Tom.

Agreed on all fronts.

I'm not sure what I was hoping to achieve with this post other than just venting and maybe some re-enforcement that I'm right in pushing back in these scenarios.

For the modelling job; I'm certain AI is basically of no help at all. If my max scripting game was stronger I'm sure that'd be the best help available. It's a lengthy process to import, model, texture, light, optimise, organise, texture bake, and finally export the individual apartment models.

Your suggestion of trying to find a good video to educate Jay is a good one. Cheers,

[Juraj]

This sounds like quite a shitty client to have ;- ) But I understand if it's your main source of income, you might not be considering dropping him. But that would be the best solution.

[louisryko]

Actually, they're one of my best clients. Jay is demanding, but he get's the best out of me.

And yes, not the main source, but certainly a major one.

I did not intend to convey that they're a bad client, rather the opposite. Jay just isn't listening to me, and - as explained - simply wants to speed things up and cut costs.

I'm looking for a silver bullet to try and sway his belief in unrealistic AI accomplishments.

[James Vella]

The only 'silver bullet' I can think of is your experience/knowledge on the topic. You obviously are using AI in your work currently where its beneficial (touch-ups/people). So personally I would suggest to Jay that it would be beneficial if he has some specific AI tools that he knows of that does this sorcery. That way you can spend some R&D time to see its just hype or could actually cut costs/time. Its a 2 way street in my opinion.

[Juraj]

Actually, they're one of my best clients.

speed things up and cut costs.

Well if that isn't the definition of amazing client than I don't know ;- )

Anyway, AI for this question can be replaced by any tool or technique, it's somewhat a red herring in my opinion.
The crux seems to be client unwilling to listen to authority, which should be you. It's communication and trust issue, everlasting problem in any relationship, business or otherwise.

I personally wouldn't want client telling me what and how to do, he came to me for that expertise. But if he would have more expertise than me, than I would be a poor fit.
Usually just one of those can apply at same time for given topic. It's up to you to decode whether your client is just micromanaging AI-hype obssesed dreamer, or whether you need to pick up the slack and integrate wider AI skillset (if that's what you want to).

But this seems to be an odd relationship. And I don't think AI is the issue.

[lupaz]

But this seems to be an odd relationship. And I don't think AI is the issue.

Stockholm syndrome? :)

[dj_buckley]

It sounds to me like Jay is trying to run your business for you.

I'd suggest to Jay that he should do the R&D into it, because from your (better) understanding, it isn't currently possible that you're aware of.  If he wants to pay you to do the R&D then great, you'll come to the same conclusion but at least you'll have been paid for your conclusion (and the hassle of having Jay mither you about it).

Personally speaking, I'd drop Jay in a heartbeat. 

[Jpjapers]

It sounds to me like Jay is under the impression that he is your supervisor rather than your client and that he owns you a little bit.

I would suggest trying very hard to diversify your income from clients a little so you dont rely on his work as much. Thats a bad position to be in quite frankly because as others have said, he should respect that you know what youre doing and have the expertise to do so. Im sure we have all had clients like this in the past and all agree that they are a nightmare.

Do you have a contract in place specifying a number of free revisions before you charge extra? In the past when a problem client has come back ive put in that sort of clause so that when they want tiny little changes i can say to them, i can absolutely do that but itll cost an extra $XXX for another round of revisions. They often realise that moving a chair 2inches to the left isnt worth it. Or they will spend the time to collate all of their feedback into a single round of revisions saving you both time.

Going beyond AI it seems the reason he wants AI to do the work is because he thinks it would save him money. So If thats what he wants then you should agree on a charge per apartment (or per sqft perhaps if theyre inconsistent), set a time limit per sqft and then work to a quality that you can achieve in that time. If he isnt happy then you explain that if he wants higher quality then you need more time.

Or you spend some time doing r&d yourself into workflows that can speed up your output to make the bulk of the work easier on yourself in the long run. Is there a modular approach you can take in any areas? Could you spend time creating some parametric assets or rooms using something like railclone and its splineID tools to position furniture against walls or for kitchen cabinet arrangements etc. Are there scripts out there to help already for things like mouldings and kitchen cabinets? These sorts of things could save you time and hassle and can be reused in future for other clients too and if you save yourself time, you can either pass that saving on to the client depending on how youre billing. Or you can get the client their work faster with less of a time cost to yourself.

If youre doing 200+ apartments its worth looking at your workflow and seeing where you can semi-automate things with some upfront work even if that r&d time is unbilled.