[HN Gopher] Playing Zork with AI-generated imagery [video]
       ___________________________________________________________________
        
       Playing Zork with AI-generated imagery [video]
        
       Author : mr_walsh
       Score  : 65 points
       Date   : 2023-01-08 17:01 UTC (5 hours ago)
        
 (HTM) web link (www.youtube.com)
 (TXT) w3m dump (www.youtube.com)
        
       | JoeAltmaier wrote:
       | Couldn't the pictures be better with a little more detail added :
       | "with a Dungeons and Dragons vibe" or "in the style of fantasy
       | artist Tony Sart"
        
       | wahnfrieden wrote:
       | I played this game on an IBM PC before I could read or type, with
       | my mother reading aloud and entering what I decided. The visuals
       | I imagined are still vivid in my mind's eye
        
       | anthk wrote:
       | I wish Dungeon (condensed pre I-III Zork) was under MIT license
       | instead of a non-commercial one.
        
       | madaxe_again wrote:
       | Considering that you can get chatGPT to accurately play zork (the
       | Linux terminal emulation route is one way), and considering that
       | folks have been experimenting with getting it to provide prompts
       | for stablediffusion, I could see this being done with a different
       | ML stack, possibly with an awful lot less palaver. They mention
       | scene continuity for instance - chatGPT understands this pretty
       | well, when I've got it to play stuff like the witness and then
       | quizzed it on what's going on - literally the example they gave,
       | using a flashlight in a dark space.
        
       | FatActor wrote:
       | > "We never got to see zork..."
       | 
       | Well... There have been so many visualizations of Zork over the
       | decades. I still default to the InvisiClues images from the early
       | 80's, which I think are far better than the first graphical
       | attempts of the white house (Zork Zero).
       | 
       | However, this is extremely cool. It's like real-time Myst (yes, I
       | know that exists, but that's a raytraced version of the original,
       | not this). The meta data optimization was unexpected.
       | 
       | The video ended too soon.
        
       | warning26 wrote:
       | Neat; anyone know if this is going to be released anywhere?
       | 
       | Would love to see this implemented generally for other text
       | adventures!
        
       | imranq wrote:
       | Very nice! I've never played Zork, but if someone is able to
       | solve for scene continuity, I can imagine this opening up a whole
       | new way of playing games.
       | 
       | Either way, there's going to be a whole new field to adapt these
       | generative models into games and productivity apps. Its going to
       | require a lot of clever thinking to avoid nonsense output and
       | UX/UI where generative imagery makes sense, but its a field ripe
       | for incredible advancements
        
         | tyingq wrote:
         | >solve for scene continuity
         | 
         | Yeah, the abrupt change in the house when you open the window
         | starting around 2:16 or so shows it's got a ways to go there.
         | https://youtu.be/ZpCrBBj6AWE?t=137
        
       | raldi wrote:
       | Zork co-creator Marc Blank worked at Google for many years, but
       | left about a decade ago. It's a shame he's not at the company for
       | this project.
        
       | CamperBob2 wrote:
       | Yeah, that was the first thing I did when I got Dall-E access.
       | Basically just fed text from Zork room descriptions into it.
       | 
       | As somebody who eagerly bought into Infocom's advertising back in
       | the day, about how the best adventure game graphics were always
       | going to come from my imagination, it's sort of disillusioning
       | how well it worked. E.g., https://i.imgur.com/xxPn3Mh.png is
       | better than what was in _my_ mind 's eye at the time,
       | notwithstanding the Escherian chair. I honestly don't know if I
       | want to watch that video.
        
         | ghaff wrote:
         | I don't know. That's mostly fairly generic imagery.
         | 
         | But a lot of Infocom puzzles did depend on you translating a
         | verbal description into the appropriate object--e.g. U-shaped
         | metal = magnet.
         | 
         | Disclaimer: Sometimes beta tester back in the day. And knew
         | lots of the people involved.
        
       | riidom wrote:
       | I don't like it. Looks as inspiring as when someone accompanies
       | each room description with the result of a quick image search.
        
         | WillPostForFood wrote:
         | I like the concept, but on an artistic scale of 1-10, this is
         | like a 1.
        
           | JoeAltmaier wrote:
           | More GPU power and the pictures might get better?
        
       | pettycashstash2 wrote:
       | I love this. Playing it text version is still the best way to let
       | your imagination run wild.
        
       | dinobones wrote:
       | This will be looked back at as an example of when Google missed
       | the beat and lost their exceptionalism.
        
       ___________________________________________________________________
       (page generated 2023-01-08 23:00 UTC)