Something EPIC is brewing.
I should be working but...
Like many others who have been sucked down the rabbit hole of AI over the last few weeks and months, I have lost countless hours to both ChatGPT, and Midjourney V5. Both pieces of software feel like an infinitely large leap in technical ability from what was possible before them.
With Midjourney v5, for the first time in my life, I'm unable to break down how I think a program works into something that feels logical. It seems like there is some sorcery happening that just defies explanation.
I asked Midjourny to create a scene full of delicious baked goods in the style of insects.
How can a ridiculously surreal scene, that only I could have dreamt up, be so perfectly rendered in seconds with such exquisite detail?
Putting things in perspective
To make matters even more confusing, despite Midjourneys obvious mastery of creating believable scenes that include perspective, 3D geometry, shadows, reflections and scale, it is clear that it knows absolutely NOTHING about the actual 3D structure of the scenes that it creates.?The 3D attributes of every image are merely an illusion, something that it has learned to fake perfectly.?I tried numerous ways to trick the AI into providing me with some 3D data,?including asking it to render a 3D grid of 25 squares containing 25 instances of the same object from different angles and generating a height map of a simple object.?All the experiments failed miserably…the AI appeared to be annoyed..like a magician when you point out how their illusion was performed…it served up distorted images or completely ignored my prompts altogether.?
Flatland...
The reason for this complete lack of 3D understanding is because Midjourny was trained on over 100 Million 2D images. These images could not teach Midjourny how to understand 3D dimensions or perspective, or scale, just how to organize different colored pixels into shapes that look like things that were described in the image tags.
The Solution...
So how would you create a 3D version of Midjourney, a piece of software into which you could describe a scene and then not only see it rendered, but also explore it on a screen, or in VR or a Voxon Photonics 3D volumetric display??
A procedurally generated world full of real and imaginary objects and characters created in an instant in any style that you can dream up. A world where every item has an X,Y and Z dimension and occupies a "volume" in the world.
领英推荐
An EPIC 3D library
You would train your AI not on a database of 2D images, but on a database of 3D geometry.?A curated library of 3D characters, animals, mountains, cars, books, tools, castles, fish, peanuts, birds, rocks, trees...... each one having known dimensions, textures, animations, mass, physical material etc.
Today I watched EPICs "State of Unreal" video in which they covered an incredible number of new technologies. It was only later this evening that I realized the significance of one particular segment of the video.?Epic announced "FAB", a consolidated library of millions of curated 3D assets that will be user-created and cross platform ( yes EPIC is even building direct compatibility for Unity ).?What most people might see as a fantastic marketplace for building 3D worlds, I now see as something much more.
Epic is building an enormous training dataset of curated 3D assets that could be used to train a "geometry- based generative adversarial neural network" (GAN). And the results will be mind boggling.?
"Are you ready to bring your wildest dreams to life? Look no further than Endjourney! Our state-of-the-art AI technology can create your very own virtual world in an instant. With Endjourney, you can build anything from an open-world adventure to a movie location or even a recreation of your own dreams. Just provide a text-based description or an image made in Midjourney, and watch as our AI brings it to life in stunning photo-realistic detail in Unreal Engine 5. Imagine exploring a living, breathing dreamscape with your friends, just like in "Ready Player One." Endjourney makes it possible. Try it out today and start your own epic journey!"?
( I asked ChatGPT to make my text sound like an AD )
This is about as close to a "Metaverse" as I think we will ever get, and it will be made not by big corporations, but by YOU, and that will be truly EPIC.
BIM Coordinator | B. Arch. Sci | ISO19650 - BIM Project Information Practitioner
1 年Esteemed Luminary of everything 3D. Your insightful journey into AI, ChatGPT, and Midjourney V5's enigmatic depths has left me captivated. As you ponder the future of 3D modeling and visualization, I share your excitement for the potential. Let's forge ahead on this grand adventure, blending creativity, technology, and humor to shape the Metaverse into a boundless realm.
Award-winning science, tech and business editor
1 年Michael Nu?ez
Software Developer for Surgical Devices with a Tech Lead Mindset & Innovation | XR Developer & Enthusiast ??| Coding by Day, Performing by Night ?? | Biomedical Engineer at ??| I'm water my friend | The Joker
1 年Loved it! I am hyped!
Chief Technology Officer - London Design Engineering UTC | Teacher of Digital Media | Intel SFI Gold Ambassador | Public Speaker | YouTube Influencer (Rebel Base Builds)
1 年Sarwar A.
Fair point, building a huge dataset of assets definitely could end up being used to train ML models for all sorts of generation. Hell, if they can just do one for UV editing I'm sure artists would love it :)