I’m not against such a law in theory, but I have many questions about how it would be implemented and enforced. First off, what exactly counts as AI generated? We are seeing more and more that AI features are being added into lots of areas, and I could certainly envision a future in few years time that nearly all photos taken with high end phones would be altered by AI in some way. After that, who exactly is responsible for ensuring that things are tagged properly? The individual who created the image? The software that may have done the AI processing? The social media site that the image was posted on? If the penalties are harsh for not attributing ai to an image, what’s to stop sites from just having a blanket disclaimer saying that ALL images on the page were generated by AI?

AI models for WHAT? There are a ton of models out there for all kinds of purposes, and sure, some of them might be trained on open data.

The Open Model Initiative https://openmodel.foundation/ is working towards creating some fully open sourced models, starting with an image generation model, though nothing yet. I don’t know the details of what kinds of data they plan to train on, but I imagine that they probably wont be training just on creative commons data, but someone could theoretically do that given that the entire thing is fully open.

There was also an image generation model called CommonCanvas https://huggingface.co/common-canvas which was trained on creative commons data.

Low res textures that would get resized with a bilinear filter, as opposed to the PS1 which used no filtering, resulting in a sharp but pixelated look.

I like swarmui. It’s an easy to use interface for generation, but it’s built on top of comfyui. The easy interface can handle most common scenarios, but if you need a really complex workflow, you just click over to another tab and you have direct access to the actual comfyui, to build up whatever workflow you want.

Headline: “This company got a copyright for an image made entirely with AI”

From the article: “In the certificate of registration, also viewed by CNET, the office said that the AI-generated components were excluded from the copyright claim.”

Clickbait aside, it’s clear that there was an element of human authorship here, and those aspects of the image are what can be copyrighted. This is in line with recent guidelines that the copyright office provided regarding AI generation.

Soulseek has had my back so many times when I need to find a really obscure track that can’t be found anywhere else.

Looks like an exciting model. I am curious to see how it will be with finetunes and controlnets. Could potentially be better than sdxl. It does unfortunately appear to be a non commercial license though, which might limit interest.

Swarmui is really awesome, I’ve switched to it as my main ui recently.

I guess I just don’t see the appeal. If I wanted to interact with an AI, I can do it right now without launching a game to do it. And pointless NPCs are already one of the most annoying parts of games for me. I don’t want to waste my time reading meaningless dialogue from 100 characters trying to find the 1 person who will actually let me advance and continue playing the game. It will be even worse when they are all taking an unlimited stream of shit instead of 1 or 2 canned responses, because it will just be even harder to know who is wasting your time.

When a bunch of RPGs start doing that, they are all just going to start feeling the same.

Well, hopefully this will at least force stability’s hand in some way and get them to at least make an official statement instead of just remaining silent.

It’s been a wait, but glad to see it finally out. Looking forward to playing with it later.

Looks interesting, but also somewhat complicated. Now generation has to run through 3 models? I thought it was bad enough having a separate refiner model for sdxl.

Wow, this looks amazing!

So is this just a new stable diffusion gui? Any notable features, or does it just work the same as all the other ones?

Saw a clip of someone working with it on an rtx 3090. Image was updating multiple times per second, and looked pretty good quality.

Well shit, they finally released the thing he was talking about a year ago.

Always have to take what he says with a grain of salt. I remember last year when he said they were a few weeks away from releasing a model that was like 30x faster.

Been trying this with my local comfy installation. It’s a lot of fun! For some reason it tells me I don’t have control net installed, even though I installed the stuff listed on the GitHub page.

Now I need to try to learn to paint in krita a little better

I feel like the issue could be largely resolved if developers would just make a lot of the downloads optional. For as long as I remember, installers for PC applications have given the user options for which parts of the applications they want to install. Yet for games these days, they just take an all or nothing approach. Let me skip the ultra textures, and all the extra choices in languages I don’t need. It seems like it should be such an easy thing to implement!