4.2 Gigabytes, Or: How to Draw Anything

Re:4.2 Gigabytes, Or: How to Draw Anything by archagon on Tuesday August 30, 2022 @03:53PM (#32654289)

This AI is not intelligent. It will not “learn” in the same way that humans do.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by Gigachad on Tuesday August 30, 2022 @07:34PM (#32656542)

What is this based on? So far these image generation AIs are learning to draw at a faster rate than I am.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by nanonomad on Tuesday August 30, 2022 @08:08PM (#32656774)

https://rom1504.github.io/clip-retrieval/?back=https%3A%2F%2...

Mostly LAION 5B, IIRC. Keep safe mode on unless you want to be inundated with 90s era pornog.

Browsing the dataset for inference terms can help build better prompts. For example, juicy seems to be a particularly weird term involving grapefruits index fingers and japanese people. so id probably use a synonym instead

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by quadcore on Tuesday August 30, 2022 @03:56PM (#32654330)

There isnt a result it wont be able to mimic in a short period of time.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by Jevon23 on Tuesday August 30, 2022 @04:19PM (#32654547)

Well it’s already been trained on millions (billions?) of images and it still has trouble with plenty of things.

Scenes with more than one person in them, or multiple people and objects interacting in complex ways, are the most obvious cases.

I’m sure the technology will keep improving but I think it’s possible to be overly optimistic as well.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by quadcore on Tuesday August 30, 2022 @04:39PM (#32654762)

Finaly a fair point. I dont know for sure and some here do, but my feeling is that we are still in a Moore's law in this art generation thing. If thats true, it means 10 years from now the AI will be able to mimic the very bests to perfection. Taking a pencil will become hopeless.

20 years from now they'll click a button and you'll get a fully randomly generated pixar movie thats as good as the originals.

Im software engineer and amateur illustrator. I have always welcomed technology. Always felt good about it. Copilot? No worries, please automate my job, if humanity dont code anymore, I dont think its gonna kill us the slightest on the contrary. Art? Mark my words, this wont go well with people souls. This is an obvious evil mistake. Im still confident some wises will stop this heresy before civilization collapses (I like to dramatize like that but still this is bad imo)

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by raindropm on Monday September 05, 2022 @10:28PM (#32732001)

Especially when you realized that those magical image AI put out comes from hard work of fellow artists that it aim to replace. I think non-artist people will ever understand this pain.

I'm also the type that welcome new technology into my workflow, always one of those early adopter, but I have a hard time this time

...or maybe I'm just overthinking (Life finds a way!, right?)

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by archagon on Tuesday August 30, 2022 @04:02PM (#32654387)

Highly unlikely, unless by “mimic” you mean “vaguely evoke”. There is no actor behind these models. Recombination is only a very limited form of intelligence.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by quadcore on Tuesday August 30, 2022 @04:08PM (#32654450)

Its also supervised in the sense that for-profit companies work on it to grab as much value as they can.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by trention on Tuesday August 30, 2022 @04:42PM (#32654787)

The fair use argument will probably get scraped when the case gets to an EU court. Then no mimicking will be legal.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by quadcore on Tuesday August 30, 2022 @05:13PM (#32655149)

Maybe we should go even further. There is no creativity in whatever that thing generate, it is always sophisticated plagiarism. Therefore training those models out of close regulated research environment and selling the output should be prohibited.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by trention on Tuesday August 30, 2022 @05:19PM (#32655225)

I actually have zero problems with that happening. Also, your opinion of your own ability to use sarcasm successfully is at the very least highly suspect.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by quadcore on Tuesday August 30, 2022 @05:34PM (#32655419)

Also, your opinion of your own ability to use sarcasm successfully is at the very least highly suspect.

Please rephrase I dont get it. Do I not sound sincere? Rest assured I absolutely am.

[ Parent | on HN ]

Re:4.2 Gigabytes, Or: How to Draw Anything by astrange on Wednesday August 31, 2022 @12:14AM (#32658300)

EU doesn’t have fair use in the first place, that’s a US concept.

StableDiffusion was trained by an academic group in Germany and they do believe it’s legally compliant though.

[ Parent | on HN ]