Interesting, is it computer vision or you just gave the image a name?
Is there a newer or any .rbxl of some example NPCs?
Thanks!
it’s computer vision, based on the content of the image, you can ask questions about it:
How did you manage to integrate computer vision into roblox?
Also where can i try this feature?
Probably not a true computer vision. It is likely get the description or the title of the decals.
Get close to the decal and ask npc what is on it
How do you implement a computer vision into roblox? since roblox limiting developers to do things like that how did you apply it? is it an api from other site?
I tried uploading an image, and it got it wrong, i uploaded an image of a purple square and green triangle and it says that it sees a beautiful sunset.
Can you explain how the decal vision works?
Its actually a computer vision
Now since you gave an additional clue like that…
It is likely the one where the AI upload the decal to LLMs that have vision capabilities. With that very poor accuracy, I have a feeling it is the smallest model to keep the costs low. If not an LLM, then it’s likely just regular text-to-image, and might be passed to LLM.
Theoretically, it’s not very hard to implement to create these kind of things. You just need to connect an API with it.
He’s keeping it quiet because he must have underestimated how easy how people can create LLM AIs and don’t want to reveal it and make his work less competitive. I could implement it without third party services, but my laptop not have enough VRAM.
I didnt know roblox can do it. So that only works with decals/images?
Or is there a way to attach a camera to npc’s head and a computer vision model can use the npc’s head camera to detect thing?
Competition is understood, even tho i dont have any intentions to bring llms to roblox for profit, i use machine learning and Llms only as a curiousity side projects for fun
No, it certainly not from Roblox. He must have an API provided by someone else that connects to that model.
Otherwise, Roblox would provide a better accuracy than this.
As for second question… Im not sure if you can actually get a screenshot of an environment in it’s purest form (data, not image)… Something from this (Ray Tracing Module).
I see, roblox basically has no computer vision, and the guy is sending the image to external ai api and the ai sends the description to roblox as a string
Yep! To show how easy to create an agent like this, I have provided 90% open source AI agent here. It doesn’t have vision capabilities yet though… So I might implement it to compete with this guy once I completed with my DataPredict Neural stuff.
Seems interesting, but tutorial on how to set it up is missing
Well it got tutorial in the documentation. If I have time, I’ll create a video tutorial instead. Would that be better to you?
1 Like
Yeah awesome, also i will check documentation