Hi guys I've just tried those products like Adobe firefly， ChatGPT and Midjourney.
I'm on a 3070ti and they can give me 512p images in less than 4 seconds, with correct prompt text and negative texts they can bring me good results.





With this small online demo, I can create 4 of this in 4 seconds, with just a few prompt text, Say if I want to do this in Houdini, or Maya, or any DCC, how many hours will I(you) use?
PS: You can try the efficiency of generative AI here
https://stablediffusionweb.com/#demo [stablediffusionweb.com]


What's your opinion on generative AI?
I've heard that Netease boss is in a hurry to push their concept artists to develop drawing AI in order to fire them.

I've asked some of my friends working in AI.
One of the NLP researchers said for sure AI is gonna make lots of artist lose their jobs because their bosses could just buy n RTX x090 card to do all the concept arts. 3D modelers will live longer, but just a little bit. Generative AI will handle 3D modeling later. And for 3D rendering? Why using it while we can generate 2D images without calculating where each of the rays the light emits and bounces by the objects? You can draw a masterpiece 2D image AI of photorealism without the knowledge of 3D.


Another data scientist Phd said that generative AI is just a lossy compression of the entire internet, what you generate is just an approximate guess of the theoretical truth, but with 80%~90% correctness. Therefore, in the future, those who can produce high quality non-AI generated materials will have their value. Moreover, since generative AI has less control in terms of strict elements like precise coordinates, those work which requires entering precise numbers will not be replaced (like modeling a specific type of a car).

What's your thought on that? Do you think traditional 3D-DCC and its workflow will be obsolete if you are an artist?

yes
artists are doomed

just like The Pentagon when quantum computers start hacking it all the time

1. Both of your friends are correct.

2. It's a bit hard to see how it's related to Houdini... if someone makes a useful AI for, say, Retopology, it's surely possible to make a plugin for Houdini. Houdini is much more like a framework than an app.

Also,

Hi guys I've just tried those products like Adobe firefly， ChatGPT and Midjourney.
I'm on a 3070ti and they can give me 512p images in less than 4 seconds, with correct prompt text and negative texts they can bring me good results.

Before you start getting worried about AI, you should probably ask your friends about how these tools actually work. Adobe Firefly, ChatGTP and Midjourney are all cloud-based. They don't use your 3070ti. You can use a $400 laptop and they'll still give you results in the same time.

There's UE5.2 presentation on ML training for deformations, it uses H for the process, very cool stuff.
You can't do "generative" AI without dataset to infer from, and H might be one of the best solutions to create such sets, be it ML deformer, image-based LoRA, or just a whacky (or impressive) line-to-sim/art converter (as seen in the ML Terrain teaser from sidefx).

@raincole is correct about cloud-based solutions. You would need a good GPU for local Stable Diffusion for example, for cloud stuff.. not so much