AI art

[CZghost]

Kind of a hot topic these days, the AI art. What do you think of an AI art? Did you use some AI art tools, like Midjourney or DALL-E? Did you like it? And if so, what kind of pictures did you generate? Feel free to also share some of your creations as well.

I've got into AI art recently, and I must say that the results are astonishing. AI has jumped miles ahead with the quality of the pictures, which is both good and bad. Good in the sense that now people can actually generate some serious art with AI tools, and bad in the sense that it's gonna be easily abused for malicious intents, more than ever. AI is better and better at generating convincing imagery, and we need better and better tools to combat those. But as a tool for artists? It's great! I've had this opinion towards AI art that it's ruining people's livelihood, but I don't actually have this opinion anymore. Creating a good looking art with AI is insanely hard if you have a specific vision that you expect the AI to follow and understand. And I've experienced it firsthand when I've started this little project of mine. It's got a codename "The Czech Lion", where I'm attempting at recreating the double-tailed lion from Bohemian coat of arms (Bohemia in Czechia) as a realistically looking animal. Of course retaining the significant feature - its two tails. And let's say the journey to the end expected result is kinda thorny. First of all, I don't have a Plus Tier of OpenAI subscription, I only have my Free Tier, which limits my usage of DALL-E v3 to 2 generations per day, and also limits my quota of attachment data (I'm thinking like couple of MB, don't remember exactly), and no, deleting attachments don't help to free the quota. Once you attach it, it is immediately uploaded, and even deleting it still counts as used quota. Too bad if you realize you're out of space and need to remove some attachments you don't really need to free up some space for attachments you actually need.

This is second image of the 1st generation:


If you know how the lion in the Bohemian coat of arms (which is Czech Republic's official small coat of arms) actually looks like, then you immediately realize what's wrong with this picture. However this one looks otherwise perfect, so that's a positive step further. The first image from this batch showed a bit of wing-like features on the shoulders and the front legs, which were a bit distracting (like Asterix had those wings on his helmet). I won't post this picture here, so I can keep it moderate a bit. There were also 2 pictures from second generation, those looked even worse, so I decided to discard those.

Then I've pointed those negatives, and of course the positives, to the AI. I've made a mockup with annotations ammended to the 2nd image from the 1st generation, highlighting its strengths and weaknesses, and painting it as a model picture for future generations. Well, DALL-E mistunderstood my prompt, and generated 4 more pictures (two original as a response to my prompt, and then it realized it fucked it up, and generated two more pictures, attached below, and it realized it fucked up again, but it was already out of quota for that day, so this is where it stopped).



Happy little accident I'd tell you. While I'm disappointed it failed to do what I wanted TWICE, I can't ignore those pictures look dope! I guess I'll generate more winged lions in the future, using those two as a reference, so I'm definitely keeping those. I like the top one more, simply because it actually looks menacing, with the powerful roar, the large wings and the horns. Almost like it's a halfbreed of a lion with a dragon? I also like the second picture, though a bit less, because it's got a few issues with the tail, and doesn't look as good. But I don't want to trash that one completely, either. One thing I wanted for my double-tailed lion, is to look a bit stronger than it looks in the 1st generation picture No.2, and this picture actually captures that perfectly. Minus the wings, and the weird tail, the muscles are actually visibly shaped. The reason I wanted that is to symbolize the strong will of our nation. I'd also ask DALL-E to come up with some variations for that one, all in 16:9 aspect ratio, so it serves as a good desktop wallpaper. But those winged lions? Yep, I'm keeping those and generating their variations as well.

Post your AI creations below, and post your experiences and thoughts for AI art in general.

[Mat Linnett]

So Avid have added a little AI spice to their already really useful Phrasefind and Scriptsync tools in their flagship video editing software, Media Composer. Both options make a single-user subscription quite expensive at a monthly cost of £74, but our company subscription means I've been able to upgrade all of our Media Composer workstations with these new features.

Effectively, it's AI assisted transcription runnning locally on the workstation

It takes around 7-8 minutes to transcribe an hour long clip, adds timecode stamps to the transciption and differentiates between speakers, which editors can then rename. This transcription can then be exported as a text file for producers and journalists to review before suggesting changes.

Even better, clicking the line of dialogue in the transcription takes you to the beginning of that section and even allows automatic marking and extraction of that entire part of the clip.
This makes editing long interviews ridiculously quick and simple.

According to one of my editors, it does occasionaly glitch out and turn to gibberish part way through some transcriptions.
But that'll get better over time, and it's still a hugely useful tool.

Better yet, it doesn't make the worker redundant, but instead makes their job easier.

[Eraser]

Mat's example of AI sounds like a useful thing, but note how it's not a process that involves creativity. It's a quantifiable thing that can be rated from good to bad based on tangible parameters.
The popular generative AI's for doing "creative" things seem largely overhyped to me.

[CZghost]

So Avid have added a little AI spice to their already really useful Phrasefind and Scriptsync tools in their flagship video editing software, Media Composer. Both options make a single-user subscription quite expensive at a monthly cost of £74, but our company subscription means I've been able to upgrade all of our Media Composer workstations with these new features.

Effectively, it's AI assisted transcription runnning locally on the workstation

It takes around 7-8 minutes to transcribe an hour long clip, adds timecode stamps to the transciption and differentiates between speakers, which editors can then rename. This transcription can then be exported as a text file for producers and journalists to review before suggesting changes.

Even better, clicking the line of dialogue in the transcription takes you to the beginning of that section and even allows automatic marking and extraction of that entire part of the clip.
This makes editing long interviews ridiculously quick and simple.

According to one of my editors, it does occasionaly glitch out and turn to gibberish part way through some transcriptions.
But that'll get better over time, and it's still a hugely useful tool.

Better yet, it doesn't make the worker redundant, but instead makes their job easier.

This is also an interesting use of AI. I've got my lifetime license for Filmora, and recent versions contain AI assisted tools, including transcription to captions (inside the video), which is super useful for shorts. Though, it doesn't support Czech language atm, so I often opt for a manual transcription.

Mat's example of AI sounds like a useful thing, but note how it's not a process that involves creativity. It's a quantifiable thing that can be rated from good to bad based on tangible parameters.
The popular generative AI's for doing "creative" things seem largely overhyped to me.

I think that the overhypeness is caused by so many people using the tools to generate cool pictures with it. But those are more than anything just curious to see what it comes up with, generate one or two pictures with it, and move on. Yes, AI art is making it easier to create art, and makes it so non-artistic people can generate some mockups to get some inspiration from, or to use as some illustration in their other projects. But to make something serious with it, it requires patience and dedication. True AI artists tweak the generative prompt and generate thousands of pictures before they find the one they like. And even then, the final generated picture is often tweaked manually in Photoshop or something similar (like tweaking exposures, fixing some bugs/artifacts in the generated image, etc.), so it's a job basically.

[Eraser]

But to make something serious with it, it requires patience and dedication. True AI artists tweak the generative prompt and generate thousands of pictures before they find the one they like. And even then, the final generated picture is often tweaked manually in Photoshop or something similar (like tweaking exposures, fixing some bugs/artifacts in the generated image, etc.), so it's a job basically.

Sooo..... what's actually won then?
Instead of spending countless hours typing in prompts and having to do manual touch up work you could also, you know, draw the image yourself and have something that much closer resembles your envisioned end result, doesn't have any of the telltale problems AI generated images still have, does actually have some soul and feeling to it, is a creative process of positive reinforcement rather than tediously correcting software generated errors, is a 100% original work and, oh yeah, doesn't infringe on copyright straight out of the box.

[CZghost]

Yeah, the copyright. I'm not sure which sources did DALL-E use, I guess Google Images might be the source, which could be problematic. But here's a neat thing about Google Images: you can actually filter images based on their licensing. That means you can actually filter out images that do not utilize a free license (like Creative Commons).

As for drawing it yourself - that implies you know how to draw. I don't. I'm not actually an artist myself, I'm trying the AI art out tho, to see if it will suit me. If not, I'm just not doing art then. Might as well just pay some artist on Fiverr. Yeah. I'd like to see the direction I can actually push DALL-E with, and see how well it does make the picture. I'm actually scared of paying and artist and then using their art, because I've heard about many of people that got scammed by the artist they paid, as the artist was claiming the art wasn't comissioned and tried to file a lawsuit against the person using the art. So I'm scared of that.

[Eraser]

Eh I dunno, learning to do art seems like much more valuable and satisfactory path to take than dicking around with prompts to get a piece of software to sort of do what you want it to do.

[CZghost]

Yeah, I guess it is. It's often said that road to hell is paved with good intentions. But the tools are there, and everyone can use them. Wether that's good or bad, doesn't really matter at this point. People have already been abusing it, it's always gonna be abused, it's a never ending war. It wouldn't be wise to remove the tools tho. They're actually quite useful. It speeds creating art up immensely, and brings art to the inexperienced. If not for anything else, it can be used to take some inspiration from.

[Transient]

Eh I dunno, learning to do art seems like much more valuable and satisfactory path to take than dicking around with prompts to get a piece of software to sort of do what you want it to do.

I use an AI image generator to give me illustrations to use as visual aides when I play D&D. It's useful for players to show everyone what their characters look like, and it's useful for me as a DM to illustrate the settings my players enter and the monsters they fight. What would normally take me days to illustrate takes mere minutes. It frees up my time to focus more on the campaign story.

As long as people use it for personal reason like in my instance, I think AI art specifically can be a net boon.

[Eraser]

That's an example I can really appreciate, Transient. I imagine it spices up your D&D campaigns quite a bit.

[seremtan]

tbqh i'm pretty much over generative AI. every image looks instantly fake and the only pleasure i get from looking at them is spotting the additional limbs/fingers, the weird skin flaps, and the oval irises

knowing that what i'm looking is basically hi-tech plagiarism doesn't help either

[Captain]

barf

Those look retarded

[Zerofactor]

I think it's garbage that sucks

[Bacon]

I think it's garbage that sucks

Holy shit. WB