Generative AI and OA

[Godd Howard]

Hey! I'm sure this has been talked about before on the forums, but it's pretty late in my timezone right now, and I don't have the energy to go searching for information on this. I felt this was probably the best place to ask this. I've noticed that many of the articles on the site contain AI generated images. My question is, are these images intended to be permanently used, or are they placeholders? Obviously it's unreasonable to expect there to be art made specifically for every article on the site, I understand why AI is used on these pages, but there are quite a few downsides with using generative AI for the art, and I would imagine there are probably some free images out there that could replace some of the AI stuff. It's not a massive problem for me, and I'd be interested to hear other perspectives on this topic, I'm pretty open minded, so if there is a good argument as to why AI images are best option for these articles, I could likely be convinced..

[DSPE]

OA is for humans to make! Putting Generative AI on pages might discourage new artists fro making stuff for them. But I don't have much of a problem with AI- generated images though.

But anyways, what downsides do you think Generative AI has?

[Worldtree]

The generated art is from the initial interest in the topic and mostly putting Keith wigdor’s experiments up on the site alongside other people.. but then we had a backlash

I was initially much more interested in generated art but now I’d vote out 90% of it - and now try to preserve art made mostly by a human.  The santarchai image is the one exception I’d always vote to keep.

So yeah they’re placeholders
If you have a better option for any of the images, feel free to propose them/ point them out to us.

[Godd Howard]

OA is for humans to make! Putting Generative AI on pages might discourage new artists fro making stuff for them. But I don't have much of a problem with AI- generated images though.

But anyways, what downsides do you think Generative AI has?

I've got a few, but my primary issue with usage of Generative AI in OA is it tends to hallucinate or depict things strangely. Generally it's just small things, for example in the Megastructural Migrations article there is an AI generated image that seems fine at first- until you notice that many of the mountains in the background are cut in half, and the issues with perspective in the foreground. I imagine this would probably become a larger issue with images of species, where the appearence of them is more important. 

As well as that, granted this is a personal opinion, but I don't know how ethical these AI models are as a whole. There already has been some issues with generative AI simply being built off blatantly stolen artwork, I don't know if it's a great idea to use these tools if there are likely free images out there that, while they may not exactly depict the content of the article, are at least somewhat related to it.

[Drashner1]

To add my 2c to the discussion:

As has been mentioned, when AI generated artwork first became generally available, there was a flurry of submissions created with it in the vein of 'oh cool! Let's use this!'. However, since then our views have evolved a good bit (OA is in a pretty much constant state of ongoing evolution rather than top-down pre-planning and evolution can be...messy). This has resulted in some images being removed and our general approach to such images being modified. Site artwork is not one of my primary foci so I won't be surprised if the following is modified by posts from other members of 'the management' who are more directly involved in that aspect of things, but in a nutshell...

1) AI generated artwork already on the site is considered to be a placeholder until we can get around to replacing it with something either fully human generated or generated by a combination of human and AI effort.

2) Currently, artwork submissions are considered/prioritized more or less as follows:

a) Human generated artwork is always preferred

b) Human artwork in combination with AI in some form (e.g., AI generated basic images/background/elements with human created art on top for main elements).

c) AI generated artwork

3) AI artwork of whatever degree must be credited appropriately. Our current guidelines on that can be found on our How to Contribute pages. More specifically on the Artwork Submissions page under Third Party Artwork.

4) All artwork submissions (as with anything else) are subject to review and consideration by the OA Editors. If we decide that a given submission is problematic for some reason, we may choose to reject it or at least relegate it to a secondary location on a given page.

Hope this helps,

Todd

[stevebowers]

I've got a few, but my primary issue with usage of Generative AI in OA is it tends to hallucinate or depict things strangely. Generally it's just small things, for example in the Megastructural Migrations article there is an AI generated image that seems fine at first- until you notice that many of the mountains in the background are cut in half, and the issues with perspective in the foreground.

It is not that simple. The Megastructural Migrations image is not just an AI image. I put it together from at least three different AI generated images, as well as a model of a Bank's Orbital which I made myself. So if there are any 'stitching errors' they are probably my own work rather than the work of the various AI generators I used.

This image replaced a previous image I made using a landscape generator and a different view of the same Bank's Orbital model; if I recall correctly I simply drew the migrating population on by hand, but it looked a bit fuzzy; so I replaced it with this one which is quite a bit clearer. In due course I may replace the image again, but I might still use a mixture of models I have made myself, found images from various sources and AI generated images.

One source for landscape images I sometimes use is Google Earth, which has an almost infinite number of views of Earthlike landscapes. I also use Space Engine, Terragen, and an addon that generates landforms in Anim8or. I also use photos I have taken myself. All these images (including the AI generated ones) are heavily edited in Gimp, so I consider them to be my own work, not simply AI images.

On the other hand Keith Wigdor's AI generated images are mostly, or completely, generated by AI programs, so they are more problematic I have more than a hundred of his images, some of which are in his gallery, but it is tricky to find uses for them in articles.

[stevebowers]

I've made a breakdown of the various components in this image, showing where they came from.
   


I can now see the splices you are referring to, but they are actually the result of errors I made in extending the image, and were not part of the original components. I'll have a go at smoothing these glitches out, if I can.

[DSPE]

I really like the image!
It looks like a real scene.

[stevebowers]

I've used some Gimp tricks to clone certain parts of the image to remove my previous stitching errors; this version is a bit closer to what I intended.    

[Godd Howard]

I've got a few, but my primary issue with usage of Generative AI in OA is it tends to hallucinate or depict things strangely. Generally it's just small things, for example in the Megastructural Migrations article there is an AI generated image that seems fine at first- until you notice that many of the mountains in the background are cut in half, and the issues with perspective in the foreground.

It is not that simple. The Megastructural Migrations image is not just an AI image. I put it together from at least three different AI generated images, as well as a model of a Bank's Orbital which I made myself. So if there are any 'stitching errors' they are probably my own work rather than the work of the various AI generators I used.

This image replaced a previous image I made using a landscape generator and a different view of the same Bank's Orbital model; if I recall correctly I simply drew the migrating population on by hand, but it looked a bit fuzzy; so I replaced it with this one which is quite a bit clearer. In due course I may replace the image again, but I might still use a mixture of models I have made myself, found images from various sources and  AI generated images.

One source for landscape images I sometimes use is Google Earth, which has an almost infinite number of views of Earthlike landscapes. I also use Space Engine, Terragen, and an addon that generates landforms in Anim8or. I also use photos I have taken myself. All these images (including the AI generated ones) are heavily edited in Gimp, so I consider them to be my own work, not simply AI images.

On the other hand Keith Wigdor's AI generated images are mostly, or completely, generated by AI programs, so they are more problematic I have more than a hundred of his images, some of which are in his gallery, but it is tricky to find uses for them in articles.

I was referring to the image depicting a person on power stilts, I honestly hadn't noticed any issues with the first, like DSPE, I really like it!