Comment by jdw64
3 hours ago
When I work as a freelancer, I get a lot of requests lately to create fake AI manipulated images for scams. Especially requests to generate fake IDs using AI. Personally, I feel that there is a need for AI watermarks on image generation models, but at the same time, if watermarks become mandatory, it would effectively kill the business viability of those models. It feels like the same problem as guns and gun control.
AFAIK the big models have watermarks that are supposed to be hard to remove. But I don’t think it’s possible on local models: not just questionable because it would prevent full open-source, but if someone discovers a way to easily remove a local model’s watermark, it will work permanently.
It’s a bit analogous to New York and California regulating 3D printers (which I disapprove). But more invasive, because local models are software, and here the danger is not guns but photos.
may I ask the reason your clients don't just generate these things themselves?
Usually, this kind of work is difficult with large models. They either get rejected by policy or don't produce stable results. So you have to use local models, but local models have inconsistent quality and the setup is complicated.
That's why clients usually hear about keywords like LoRA somewhere and say, 'I heard you can generate it with this method.' But in reality, they don't know how to install a local model, serve it, run it, or stabilize the results. And that's where freelancers like me come in.
As a freelancer, it's hard to completely ignore this trend. Because freelancers like me aren't inside a company, we're usually a step behind on the latest news. So we end up searching through the information and keywords that clients share with us, organizing it, and following industry trends that we pick up when we visit companies for consulting.
In any case, most clients aren't completely clueless—they usually provide keywords like 'I heard you can do it this way' and make their reques