Hacker News new | past | comments | ask | show | jobs | submit login

I'm more worried about us becoming numb to visual novelty.

I was tinkering with Stable Diffusion yesterday to come up with ideas for an apartment interior design. These aren't even cherry picked, you can generate images like this, one every 10 seconds or so, for as long as you like:

https://imgur.com/a/mczYfnv




> numb to visual novelty

I wonder if this isn't already the case.

If I look at what people are sharing on various platforms, people share images all the time, but it's rare that an image will pop up merely for its aesthetic qualities, (except for self-promo stuff).

I can only guess, but I don't believe the AI-generated images you linked to would be shared widely, if they didn't have the story "Look what AI can do," behind them.

And maybe that's always been the case. I'm reminded of the working poor going to the Louvre in Zola's "Assommoire". They mostly come away not understanding what the big deal is. It's the "cultivated" person – who knows the stories behind the paintings - who comes away moved by them.


> I'm more worried about us becoming numb to visual novelty

Given how people are becoming numb to things like porn and violence, seeking out the ever more extreme, I'd say there's a pretty good change that you're right.

As with porn for instance, where some are seeking out amateur porn, going in the other direction, there might also be a small segment who starts to seek out ever more authentic art. That is until someone finds out that you just slap "authentic" or "Human-made" on the AI stuff.


Wow, can you explain how you did it? I tried to write ,,nice apartment'' HuggingFace link from stable diffusion, but I got this error: ,,This application is too busy! Try again soon''


There are a number of ways to do it, this is is where my images came from:

https://beta.dreamstudio.ai/

With a prompt like this:

"ultra modern apartment, leather furniture, bold dark colors, built-in fireplace"

Add modifiers and other hints to see how it influences the outcome. You can add styles too (eg midcentury modern, shabby chic, etc)

Tinker with the settings on the right (note that most of them will burn more credits per image as you move from the default).

I've tried it with DALL-E and Midjourney and like the output of Stable Diffusion the best (specifically for this type of image, they are all good at some things and bad at others)


I jumped off your prompt and went for "house floor plan" and the results were surprisingly functional at first glance. The words were garbled pixels of course, but the arrangement of lines and layouts was clean, orderly, and mostly functional.


Cool, thanks! It's a lot of fun to play with for sure


Try adding your favorite team or holiday at the beginning of the prompt.


wow you weren't kidding, I copy/pasted your exact prompt, selected 9 images to be generated, and all of them look real: https://i.imgur.com/62mIYCb.png


In the stable diffusion announcement post there is a link to a CoLab notebook that’s easy to use


That's a lot of black couches.


I kept tinkering with modifiers and it started going into a strange area.

Here's some black leather on a couch - https://imgur.com/a/7szq07v

(not exactly nsfw, but on the way lol)

Edit: Also I'm obsessed with this couch. If I were a richer man I'd have one made - https://imgur.com/a/NO4yfkY


That’s quite the conversation piece! I have been delighted by these recent models’s ability to make images that are highly interesting and engaging. That’s not an interesting AI couch, it’s an interesting couch, full stop.


I would never live in one of your apartments


My disappointment is immeasurable and my day is ruined.


:) I'm sorry!




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: