Using ChatGPT and GPT-4 to Generate 3D Content in Omniverse
HTML-код
- Опубликовано: 29 мар 2023
- See how APIs for GPT-4 and NVIDIA DeepSearch were combined to rapidly generate 3D objects with simple, text-based prompts in this experimental project from the Omniverse team.
To experiment with the AI Room Generator Extension Sample in Omniverse, access the code on GitHub: github.com/NVIDIA-Omniverse/k...
To learn more about how the extension was developed, read our Medium blog: bit.ly/42KYZ3n
Download Omniverse today to start experimenting with generative AI: bit.ly/3ZiGqAI
------------------------------------------------------------------------------
NVIDIA Omniverse is a scalable, end-to-end platform for building and operating metaverse applications. Based on Universal Scene Description (USD), Omniverse fundamentally transforms complex 3D workflows, allowing individuals and teams to connect and customize 3D pipelines and simulate large-scale, physically accurate virtual worlds for industrial and scientific use cases.
Download Omniverse today → nvda.ws/3gNZsu0
Try Omniverse Enterprise free for 30 days → nvda.ws/30uI0aq
Join the Community → nvda.ws/3ZMfc6e
Resources:
Documentation → nvda.ws/3J6xZlX
Forum → nvda.ws/3ZCQWng
Tutorials → nvda.ws/2WluMH2
Connect with us on social:
Twitter → nvda.ws/3ytuGQO
Instagram → nvda.ws/3l5hizu
LinkedIn → nvda.ws/3FbYMw3
Discord → nvda.ws/3ZSViGO
#3d
#metaverse
#openusd
#ai
#chatgpt
#gpt4 
Развлечения
This is awesome!
Maybe I'll try this if I get a new GPU
I wonder how long we wait for "text-to-edit" objects in texture or vertex level
This really reminds me of what Promethean AI was doing and the demo their showed in 2021. Great stuff!
Same! I wasn't able to use Promethean at the time so I'm really excited about NVIDIA's version.
Yeah there's a LOT missing from this video. I've gone through the blog (and all of the links) and there's no extension for this. There's also no code to create the extension. So once you get the area created with the curve, you need the Room Generation window. This is where this falls down as it doesn't exist by default in Omniverse (2022.3.3), there's no AI extension I can find, it's not anywhere. We NEED information on how to get this setup?
I'm Making a tutorial of how to use the tool :D
@@RobertoCharreton Where will you be posting that tutorial?
@@JohnLewis-old Here in my channel!
The GitHub link has been added in the above RUclips description
@@NVIDIAOmniverse how to use the github likn
This is amazing!!!
Awesome 😊
when you say "generate" do you mean generate 3d model from 0 ?? or just placing an already exicted 3d models in the library ?
How is the orientation (rotation) of the assests determined?
In this video the chairs face each other. At what step does this happen?
how do we enable the Generate Room panel?
AMAZING!!!!
This is Amazing Tools .. :D
🎉 this is magical
This is soooooo amazing! I have 2 questions:
1. How do I define a custom furniture categories? like I need to use my own old and broken sofas. Maybe the category name is "Old broken sofa"
2. How do I use it in UE5?
you basically can connect your UE5 scene to USD format and receive changes in both ways
Also, isn’t it required that you have your own collection of correctly named assets in repository that the script run by GPT 4 can pull from? I’m unsure, where these assets are coming from because I know the NVIDIA 3D model generation is a thing but not at that fidelity or integrated into omniverse yet?
that's a great question! I wanna know the same!
can I export the area I have cut as individual usd file?
Hmm! interesting! looks so very productive! I'm thinking I can make a ''basic'' layout. then send all the models through
texturing & physics, then bring it all back in. & have a finished & complete level!! beautiful !! wow! this is great!!
it would be great to see CG to finally leap off the ground!! good luck all !!
Great stuff but how to enable deepsearch? Unable to find any thorough tutorial.
How to get this Omniverse with ChatGPT service ? already in the latest version ?
Is there an API to run this on my own apps? I am interested in passing a 2D image turning into a 3D image adding furniture and parsing it back to 2D. Any suggestions?
Awesome! The placed objects are loaded from a Nucleus repository too?
Yes, that's right. Deep Search scans the assets in the Nucleus and reports the results. In this case we have it searching against assets NVIDIA has created or provides as samples
@@tjgalda1284 What sort of information do these assets come with? It's interesting to see the chairs for example, oriented in a way that makes sense, relative to the tables - so there's clearly more than just the 'bounds' of the objects being used. As far as I'm aware GPT-4 doesn't use vision/image input yet so is this meta data built into the assets themselves?
@@Arctan95 my thoughts exactly.
God i didn't expect this much growth in ai! It is amazing
there really needs to be a step-by-step video on installing this because i can't get deepsearch to recognize my local nucleus. it's not showing anything so chat gpt only loads cubes.
What is the hardware requirements?
MINE!!
i can't find generate room window. perhaps you should explain all from begining, i couldnt find in tutorials before. How to arrive to the beginning of this tutorial?
I noticed that with a lot of Nvidia’s tutorials. Lots of videos assume that you already know how to get there, and it is impossible to find where the beginning might be.
me too....having same problem
How to install generate room?
The loading bar just runs forever, anyone had it working properly yet?
Ace!!!
holy sh*t
Unfuckingbelievable, that’s it I’m sold.
only for linux and windows, seriously??
Can you name a Mac that uses an NVIDIA graphics card or any other graphics card for that matter?...
Macs blow once again...
I don't see what's so amazing. All your doing is adding furniture from a catalogue. 3d companies have better quality models and more in trend. If anthing I think this slows the process down. Way quicker to select from your own catalogue so you know what your placing in the model.
this is also ran software, NVDIA should do what it is good at, making gpu's
Can you name a software or tool that can generate a random, reasonable 3D room within a minute?