One step further forward than my attempts at 2D AI-generated art is this, which takes me back to my original computer-generated artistic playground, the creation of 3D models.


Incredible new AI 3D model generator
could transform character design


(Image credit: VFusion3D)

Generative AI has long since left the realm of only being able to handle still images, with terrifyingly realistic animations and videos emerging from simple text prompts by the day. And now, researchers from Meta and the University of Oxford have unveiled a powerful new tool capable of transforming text prompts into detailed 3D renders.

VFusion3D sidesteps the issue of a "limited availability" of 3D data to create truly impressive models, and web users are already speculating that it has the potential to transform the process of character design.



Titled 'VFusion3D: Learning Scalable 3D Generative Models from Video Diffusion Models', a paper by researchers Junlin Han, Filippos Kokkinos and Philip Torr describes how the new model is capable of "building scalable 3D generative models utilising pre-trained video diffusion models".


A flat image (left) can become a 3D model (right) (Image credit: VFusion3D)

According to VentureBeat, the team "fine-tuned an existing video AI model to produce multi-view video sequences", which teaches it to view objects from multiple angles. The results speak for themselves – the paper includes several examples of still images transformed into 3D objects, with the AI filling in the gaps with remarkable precision.

"The primary obstacle in developing foundation 3D generative models is the limited availability of 3D data. Unlike images, texts, or videos, 3D data are not readily accessible and are difficult to acquire. This results in a significant disparity in scale compared to the vast quantities of other types of data" reads the introduction to the paper. "To address this issue, we propose using a video diffusion model, trained with extensive volumes of text, images, and videos, as a knowledge source for 3D data. By unlocking its multi-view generative capabilities through fine-tuning, we generate a large-scale synthetic multi-view dataset to train a feed-forward 3D generative model. The proposed model, VFusion3D, trained on nearly 3M synthetic multi-view data, can generate a 3D asset from a single image in seconds and achieves superior performance when compared to current SOTA feed-forward 3D generative models, with users preferring our results over 90% of the time."

And you can even try out VFusion3D yourself. A publicly available demo on Hugging Face lets you input your own images, or select from a bunch of preexisting ones including Pikachu and Baby Yoda.

From the terrifyingly realistic viral Flux images to those AI gymnastic videos, {Definiterly worth watching } AI-generated content is either getting more realistic or more terrifying – and often both. But hey, don't worry, recent reports suggest AI generators could end up being their own demise.

SOURCE

I've already got software that allows me to build 3D models and output them in a form accepted by 3D printers, Mike.  I've had that for years, but I could never justify the purchase of a 3D printer.  TBH it would just be a very expensive toy, and I've probably got enough of those already. 

he so will.

You're not wrong, Robert.  As far as 2D is concerned there is a new standard named FLUX which is even more realistic, and freely available.  Sadly, the software I use is not compatible, but most of the major commercial tools (I use a freebee) are releasing new versions that will work with it already.

I literally thought this woman is gorgeous but her movement is just mesmerizing but now I'm scared because I don't know if its AI or not.



On a sidente: If it's AI, I wonder if something like this can be created with a prompt that can alter which drum beat or section of ryrhm the dancers feet would land on.

Welcome to Generative AI.  Let us turn your pleasure into paranoia.  Is it real?  Well, it's certainly not Memorex, unless we make it Memorex.

Of course, AI image generation is far from being all about replicating reality.  Aussies will (hopefully?) recognise the legendary "Drop Bear" below, about to claim the life of another tourist. 



I must admit, I'm quite tempted to turn this little fella into a model because, as far as I can tell, one doesnt exist yet.

The reason why I reinvented my model, by the way, is to make it much simpler for people (and me) to create images of this sort with her...

I am completely lost.