Artificial Intelligence (AI) generated images and renderings - Midjourney, DALL-E

That will just create an extra dimension of gibberish. Likely unusable in production.

https://youtu.be/rvHfaZIZsug

This was doing the rounds recently. Then they were caught posting ads in India for people who could knock out simple 3D models in 15 mins. Presumably that's how long 'the cloud' takes to process things.

Think my job's safe enough in the short term at least.

Let's hope it's gibberish for long enough so we can make it to retirement.

The things Nvidia and the like are doing recently with interpretation of single images and videos in to 3D objects/environments, it's got to be a small matter of time until text to 3D is commonplace.

How editable (and useful) that'll be, I imagine will be along the same lines as the images, as in the end product will just be a complete thing, rather than part of a thing that's easily processed in to other things i.e a direct line from concept to product.

Hope I'm not being naive in thinking that creating topology such as this will be beyond AI for the forseeable...

Well retopo doesn't require AI. There's obviously some high level decision making that's going on there in terms of where the density needs to lie that is probably a fair way off being automated. I'm not an expert on that front though, only played with 3DSMax's retopo that imagine probably isn't the best available.

I shot some close-up, focus stacked and retouched images for a watch client a while back, they just got back to me for 2 more shots as the renders they are using on the website are “not up to the quality of your shots”.
They actually look ok and pass muster but there is something not quite right with them, no idea of the resolution of the rendered images but i’m supplying 300+MB tiffs with no interpolation going on.
I too hope to make it to retirement and then take up art (with real pencils/paint)

That said the Nvidia Nerfs or whatever they're called seem to create more of a point cloud rather than a mesh, so yeah, not editable, or retopo-able or very useful in their current state. I'd bet they're pumping some money in to making that happen though!

Unless they're deliberately added, renders lack the imperfections - however slight - that you get with a physical product and camera lens.

Also, lighting in CG is reasonably analogous to photography but lacks much of the subtlety unless the artist really knows what they're doing.

DALL-E now open to all:

https://openai.com/blog/dall-e-now-available-without-waitlist/

.

3 Attachments

Craiyon for an easy app or webpage, my daughter loves it.

1 Attachment

https://medium.com/mlearning-ai/an-advanced-guide-to-writing-prompts-for-midjourney-text-to-image-aa12a1e33b6

just installed nvidia canvas - basically it lets you scribble with various colours which translate to different materials (sky, mountains, sand,trees etc.)
this is from a quick 5 minute play around.

2 Attachments

and the same scribble with a couple of the different render styles available applied

2 Attachments

That's pretty cool

Anyone remember Vista on the Amiga? You used to do a similar thing and after 45 minutes or so it would render a landscape. It was pretty amazing at the time.

yeah, I think so.
this Nvidia thing is wild. instant feedback from every brush stroke.

as it happens, I'm just sat having a beer waiting for a talk to start on the state of creative ai at the Leicester ai art festival.

I'll report back if there's anything insightful.

1 Attachment

There's an equirectangular version too, for 360 images you can use as environment backgrounds.

https://twitter.com/CountereCulture/status/1582418983701069826?s=20&t=IlTgwBzEWZbKg1VjsbXRjw

https://youtu.be/7KX1YqICGaE

Clever!

machine learning

In the hands of professional artists, this is a taste of what you get
https://www.theguardian.com/artanddesign/2022/dec/01/six-leading-british-artists-making-art-with-ai