Photographer Tried to Get His Photos Removed from AI. He Got an Invoice Instead | Hacker News Reader

Photographer Tried to Get His Photos Removed from AI. He Got an Invoice Instead

vice.com

32 pointsisaacfrond13 comments

He asked people to remove links to public available resource and did not have any standing for a copyright claim. He was informed of that and warned that attempts of legal actions without grounds would be billable. He did it anyway.

Almost like the real story here was FAFO, not AI.

Let's say I did a survey of images on the internet, using a crawler to look for and download each image. I make counts of how many of each file format I find, their sizes and color depths, and a hash to count dupes. I don't save or redistribute any image. I let the public buy copies of the statistics I have compiled.

Can a copyright owner claim infringement, and demand that I update my statistics to make it as if my crawler didn't find his images?

Also, your resulting data-set would be at least an order of magnitude larger than a Stable Diffusion model.

Also, I believe this is irrelevant.

Pretty much! To the actual case at hand it probably is irrelevant.

People do use the LAION dataset for something of course. Originally maybe just for image recognition models, and now of course for image generation models.

This is disingenuous. Large models are observed to reproduce works from their training data, sometimes verbatim, often close enough to constitute a derived work. You scraping metadata would never be able to do that.

The dataset in question is not a model, and contains no images; it is a long list of URLs, with associated tags. Someone who wished to train a model could use this information to download a lot of pictures from various places on the internet, then create training and validation sets; but LAION themselves do not provide images or model weights. The photographer went after the wrong target here; if he wants his images taken off the internet he should contact whoever is hosting them at some publicly accessible URL.

A model being able to produce an infringing work doesn't mean the model is an infringing work.

I can produce work myself that is infringing or not infringing as a result of learning.

I don't think your questioning of my sincerity is helpful.

I agree that generative models have been observed to reproduce training input verbatim, but that's not what I understand to be at issue in the article. The issue here is whether you can presume the consumer of a list will do something illegal with an item on the list without any evidence, and therefore threaten the compiler of the list.

Less interesting to me then the photographers case - which seems admittedly on shaky ground due to the public, third party origin of the material - is the ability to invoice people for legal charges. I love this mechanic.

Is this common across all segments of law in Germany and is there a streamlined process to reclaim legal fees for spurious lawsuits?

No. It is anchored in the right to be forgotten and the right to claim your stored PII data regarding the DSGVO to protect companies from being flooded by requests from a malicious actor or even a competitor. Without that, a company could theoretically forced out of business by the work/cost the requests generate because not answering a request is by itself a violation.

Thank you!

Reply on news.ycombinator.com