"But the most effective prompt? In terms of producing a realistic but dramatically lit landscape with recognizable mountains and hills and (okay not sheep)?"
These are the kind of comments that make me click through to the "favorite" feature in HN so that I can find them for reference later. Thanks for your effort to share this.
You can even submit your domain to the list [1]! This is useful when your domain has user-alloted subdomains and you want to enforce a stronger subdomain isolation. Ghost.io for example is there because the Ghost Foundation has sent a submission.
The infancy period of this technology is fascinating.
Think about computer graphics 15 years ago. Beowulf came out in 2007, and was developed in the preceding years- let's call it 15+ years old. And it was right there in the uncanny valley where it didn't look real, but it looks realistic. It was interesting visually, but my brain told me "this isn't correct".
And now some modern game engines are doing more realistic rendering than that in real-time.
Now look at these generative models. Some state of the art ones with humans helping are pretty convincing, but it's slow work. The more general ones like these are making these wonderfully interesting images that our brains immediately say "That's not correct".
But where will this technology be in another 15 years? I think the possibilities for entertainment are really interesting. Imagine a D&D game where the GM is vocally telling the AI what to generate, then making small tweaks, and the players are seeing the results.
Just yesterday I was asked if I was worried about losing my job because of AI and I smugly replied that us programmers will be needed even more, as interpreters. This is an excellent article as explanation I'll be sharing!
Another perspective may be - similar to how people were saying the post will go bust when internet came in with email etc. - but it turned out to be the opposite.
Maybe intelligence will simply become commoditized and programming/creating things with it will require even more programmers.
Well, isn’t it? Except we’re at Middle Ages/ early Renaissance rates of computer science and programming literacy. “The future is already here, it’s just not widely distributed yet” and all that.
I think that's what was meant by "Middle Ages/ early Renaissance" in the comment above. In this time period, literacy was a rare advantage. For instance, in early modern England, literate people were not subject to ordinary criminal courts on a first offense.
VQGAN+CLIP seem to have this dream-like quality where they generate images that are evocative of your prompts but don't actually picture them.
I find it fascinating because in some cases it's not as obvious as "lump of white fluffy matter" = "sheep" but it still manages to evoke the prompt into our brains.
I'll sometimes get an unrecognizable blob but quickly asking my SO "what is this?" she will get it... unless she consciously looks at it!
>quickly asking my SO "what is this?" she will get it... unless she consciously looks at it!
It does make you wonder about hypothetical artificial neural network-like "subconscious" layers and how "more conscious" prefrontal cortex layers potentially adjust predictions and perceptions based on their inputs. (Probably a convenient "just-so" "clockwork universe"-esque narrative unsupported by neuroscience research, though.)
It's definitely "injective but not bijective" or something like that.
Like I look at the prompt "sheep grazing on a hillside by tim burton". I look at the pic. Brain goes, "yup, that checks out". You wouldn't necessarily derive the domain from the range (preimage attack), but I can readily say, "if I fell asleep after watching Wallace and Gromit - Close Shave, and Nightmare Before Christmas, this is what I would dream".
Or it may be that the words connect everyone in their own domain. If a marketer reads it's kinda looks like a pitch, if a programmer reads it's kinda client, if a husband reads its like way to ask his wife, so on and so forth, Cleaver use of words.
The underlying problem these elaborate prompts seem to solve is that the internet contains many pictures, few of which look very beautiful.
If you look at all internet pictures of sheep, many of them will not be very exciting and depict a low contrast sheep in a foggy landscape.
So to get a picture with strong saturation and clear lines, it helps to put text there that is usually associated with pictures that have these ... like "HD wallpaper" or "made with unreal engine". Most "wallpapers" might be of dubious artistic quality, but muted colors and a lack of saturation will generally not be their problem.
This is of course not the only problem with the model. It doesn't even produce a clear image of a sheep .... but that will probably get better with larger models and more training. Similarly it doesn't seem to have a sense of overall composition and tends towards fractal or tiling-like images. But those problems are probably orthogonal to the fact that the model doesn't per se try to make good pictures ... just average ones for the description you give it.
I played around with these notebooks a while back, and wondered what you get if you jointly optimize for several different prompts. Has anyone tried this? (Or is this what the article is about?)
I honestly thought it would be about not being an asshole to people doing their jobs, and instead how much better asking nicely is. Both for treating them as people, but also for getting what you want.
So I'd say it's pretty misleading.
But with the top comment making it clear what the context is, I'm fine with it after all.
Philosophic words have been a part of programmers life sometimes but not misleading. Go in depth and you would get the meaning. I must say it is cleaver use of words.
HN just prefers the original title to avoid clickbait. The drawback is that sometimes context gets lost. On the bright side this sometimes will make you discover interesting articles which you wouldn't have checked out otherwise.
It provokes curiosity, I guess. It's like good ol' clickbait in that regard: ask a question with the promise of an answer, except the question is usually not explicit, just assumed by the user. Every time I see a contextless title I wonder "Hmm, what could this be about?" and click.
