←back to thread

2093 points pabs3 | 1 comments | | HN request time: 0.601s | source
Show context
bravura ◴[] No.42135617[source]
I made an eink rpi display for staying in touch with my parents, inspired by the poetry clock (https://www.theverge.com/23669343/ai-clock-chatgpt-poems-rhy...).

My dad didn't like poetry clock, but he does like image gen. So we got a (color) Inky Impression 7.3 and hooked it up to an RPi.

I made a basic telegram bot that you could send a verbal prompt to ("snowy day"). It would then ask which of your favorite artist styles it should create an image in. I found that presenting a list of two styles combined had cooler results. The prompt would be used to fetch a random quote on the topic, and quote and style would then be feed to stable diffusion, and maybe 30 seconds later you have fresh art and a quote on the display.

My dad then asked if we just could forward images directly there. He prefers, each day, to post an image of whatever the day is (November 13 is "World Kindness Day") and occasionally share a family photo. My mom looks forward to seeing what day he picks every day.

replies(1): >>42140564 #
1. HansardExpert ◴[] No.42140564[source]
> inspired by the poetry clock (https://www.theverge.com/23669343/ai-clock-chatgpt-poems-rhy...).

That's fun. Although, from the article:

> There’s one other problem, though. It’s well known that AI language models like ChatGPT have a tendency to make up data (sometimes known as “hallucinations”), and it turns out that’s true even if you’re just telling the time. Roughly once every 15 minutes, says Webb, the clock will simply lie about the time just to make a certain rhyme work. “The fibbing is hilarious. Sometimes you can’t tell — it might say ‘one past two’ when it’s actually ‘two past one,’” he says. He says this will be fixable but, for now, is a fun quirk of the system. “Clockwork means you get precision drift; AI-work means you get hallucination drift.”

;)