NVIDIA's Latest AI Software Turns Rough Doodles Into Realistic Landscapes (theverge.com) 35
An anonymous reader quotes a report from The Verge: AI is going to be huge for artists, and the latest demonstration comes from Nvidia, which has built prototype software that turns doodles into realistic landscapes. Using a type of AI model known as a generative adversarial network (GAN), the software gives users what Nvidia is calling a "smart paint brush." This means someone can make a very basic outline of a scene (drawing, say, a tree on a hill) before filling in their rough sketch with natural textures like grass, clouds, forests, or rocks. The results are not quite photorealistic, but they're impressive all the same. The software generates AI landscapes instantly, and it's surprisingly intuitive. For example, when a user draws a tree and then a pool of water underneath it, the model adds the tree's reflection to the pool. Nvidia didn't say if it has any plans to turn the software into an actual product, but it suggests that tools like this could help "everyone from architects and urban planners to landscape designers and game developers" in the future. The company has published a video showing off the imagery it handles particularly well.
im not overly impressed (Score:1)
for all of thoses out there, there is something similar already. https://github.com/alexjc/neural-doodle
nvidias is a bit more advanced (uses a diff net type) and has a UI, but its been done..
Got one for CAD? (Score:2)
I could really use something like this for CAD.
Where I draw what looks roughly like a cylinder and it says "do you a want a cylinder?" Even the "easy" CAD programs seem to have quite learning curve.
Example: Cylinders are very common, as are circles. Why don't any of them have a cylinder button? Why do I have to create an oval, set height and width equal to make it a circle, then extrude it?
They should really be doing... (Score:4, Interesting)
... animation AI for 2D so you can get the hand drawn look artists want. Given that hand drawn animation is a lot simpler in that only the necessary details for the thing being drawin to read well are usually drawn. I have no idea why they'd use photo realism given that many artist want to create from their imagination. It'd be a lot cooler if they perfected 2D animation line art first for traditional cartoons and anime so they'd actually be saving animators tonnes of time and money.
Animators still have a really hard time scaling and animating things we made computers to help with that but we're still stupid ass apes that can't take advantage of all thise CPU power and put it to good use given our natural mathematical dumbness.
Re: (Score:1)
so you can get the hand drawn look *SOME* artists want.
It It depends on what they're after. If this is a step towards generating landscapes for things like green-screen backgrounds and games then you don't want 2D.
Re:They should really be doing... (Score:4, Interesting)
I have no idea why they'd use photo realism
Probably because there's a much larger dataset of them.
Re: (Score:2)
I see this type of technology akin to MIDI. A 3d Model normally takes a lot of space for instructions and a lot of the work is repetitive, and a lot of detail work, that will take a lot of storage and time, work on.
Just like how MIDI, has the instrument sounds already programmed, and we can just pass the note of the song to the MIDI player, meaning you have have a full score in Kilobytes of data, vs. Megabtyes per minute.
This demo, is using life like textures, because a 2d version will very from the artiest
give me 3d (Score:1)
so I can make landscapes games!
Re: Well ... (Score:2)
Guess nobody else remembers that show. Kid made drawings and they came alive.
How that's "flame bait" is puzzling ...
Re: Dumb (Score:2)
Shape recognition has been around for 40 years now.
Your numerous detractors may disagree... but there have been times that I've actually seen you write something intelligent.
This was not one of those times.
Re: (Score:2, Insightful)
It's not just filling it with a texture. It's generating a complex shape from the concept of a tree. Draw and fill a hundred trees and every one will be different, and the branches of the tree will grow to match the shape of the tree you drew. And it will also blend it into the picture and take into account lighting, shadows and reflections. That's a lot more impressive than a texture fill.
Re: (Score:2)
there is also aspects of depth and how the texture would function in the context, I was impressed to see the waterfall rendered vs some weird river, And the tree being placed on the right spot on the hill vs just cutting down into the edge.
MSpaint is nice and all (Score:2)
But what does it do when you feed it really good pixel art, or feed it an actual landscape?
Re: (Score:2)
being that it replaced a handful of colors, I expect colors that it doesn't recognize will be ignored or replaced.
Huge for artists? (Score:2, Insightful)
It's not going to be huge for artists, it's going to eliminate them. With this technology suddenly everyone will be an "artist".
Re: (Score:2)
Everyone can be an "artist" but there'll still be a skill to using it to generate attractive art
You could train a network to generate attractive art. That's not even really hard.
Looks great in the video (Score:2)
Re: (Score:1)
Bet it looks terrible close-up
Agreed.
But give it time.
Re: (Score:2)
This stupid program is nothing more than that "Bruce 2" software that allowed unskilled morons (including myself) to click a button and claim that they had "made" a 3D mountain with impressive (and identical-looking) surroundings.
Commenter meant Bryce [wikipedia.org], which hasn't been updated in years, but is still being sold by an 'old software aggregator.'
How is this AI? (Score:2)