So pleased about AI producing portraits that were previously in short supply: and good ones too.

[WiseGrimwald]

Female dwarf portraits were so hard to come by in the past. From images like this, it is so easy to make portraits for Baldur's Gate and the like.

I just hope that the artists who drew the originals are getting full recognition. That is something that will be so difficult to check as an image that has been modified might not look anything like the original. AI can change facial expressions, hair and skin colour. It can change the background easily too.

This image is from the webpage below it.

Fierce Female Dwarf in the Forge: A Tribute to the Fallen | AI Art Generator | Easy-Peasy.AI

Discover the powerful story of a dwarven avenger. Generated by AI.

easy-peasy.ai

 

[O_Bruce]

I just hope that the artists who drew the originals are getting full recognition.

I guarantee you they aren't getting any recognition.

 

[WiseGrimwald]

That's a shame.

 

[shmity72]

AI isn't the enemy as are not guns. It's the hands behind the wheels, because well, people can be greedy. Nice portrait.

 

[OrlonKronsteen]

AI isn't the enemy as are not guns. It's the hands behind the wheels, because well, people can be greedy. Nice portrait.

This isn’t actually the case. The greed, corruption, theft, pollution, energy waste, and undermining of labour are literally baked in to the “AI” development and business models. The LLM phenomenon doesn’t exist without them.

 

[WiseGrimwald]

Not knowing what an LLM was, I Googled it and am no wiser. The Wiki page that I arrived at was written in gobbledegook, a language that I have never learnt.

 

[Skatan]

Had a bunch of female dwarves saved on Pinterest (and many, many others) but deleted my account just the other day after being fed up on the ridiculous amount of ads. But yeah, there wasn't *that* many to choose from.

Been using my own picture to make some pics to use in games lately. I dig it. I know there is a huge wave of anti-AI all over, but I am not one of those.

One is a 18th century guy and one is a 16th century guy. Though I did ask the ai to make them in the style of baldur's gate, heh. Didn't really nail that part.

 

[WarChiefZeke]

I have used AI for a lot of DnD art. If I like a character i'm playing in Tabletop, i'll get a good portrait for their character sheet.

It took some time and effort, but i've had excellent halflings and gnomes be created, who look visually distinct from one another, including sub-races like the snviferblin.

 

[WiseGrimwald]

I'll be using this one in the future. I removed my glasses using AI just this week.

Someone had made this decades ago for me, and AI certainly improved it.

 

[Antimatter]

That one definitely looks good!

 

[WiseGrimwald]

That one definitely looks good!

AI can certainly be beneficial. It depends how it's used.

 

[WiseGrimwald]

One problem with AI is that it is NOT intelligent.

Here is a dwarf wielding supposedly a halberd. -------------------------------- After telling it to replace the axe with a long handled halberd, I got this result.

Clearly AI doesn't know what a halberd is!!! It clearly thinks that axe and halberd are synonyms!!

Both are excellent images. It's just that a halberd isn't an axe.

 

[WarChiefZeke]

To be fair, the image to the right is basically a perfect replica of the standard halberd from NWN2.

 

[WiseGrimwald]

To be fair, the image to the right is basically a perfect replica of the standard halberd from NWN2.

I am perfectly aware that the second one is a decent halberdier.

The first one was my initial request for a halberd which it clearly is not.
To get a halberd, I had to specifically ask for long handles.
This version of AI clearly didn't know that halberds ALWAYS have long handles. No doubt in time the programming will improve.

 

[BelgarathMTH]

I am perfectly aware that the second one is a decent halberdier.

The first one was my initial request for a halberd which it clearly is not.
To get a halberd, I had to specifically ask for long handles.
This version of AI clearly didn't know that halberds ALWAYS have long handles. No doubt in time the programming will improve.

I'm finding through conversations with my browser AI that anything and everything related to gaming is a weak area that is prone to making it have hallucinations, and I think I understand why, at least in part.

As you know, it works by sorting through everything it finds online and identifying patterns and syntheses, then zeroing in on a kind of "best fit" for whatever question or task is put before it.

When it comes to gaming, there is a constant shift over time as games get changed throughout their development. There is so much difference between the "1.0" version of a game and whatever is the current version, that the final version of a game and what it was initially might as well be different games completely.

What was "the right answer" in say, 2016, can be completely "the wrong answer" by 2026. Trying to program AI to deal with subjects that change so quickly probably presents a lot of problems, which are exacerbated by the fact that I as the user might be asking specifically for information about an older, outdated version of the thing, where another user is asking the same question about the current version of the thing.

I find that it helps a little bit if I include version information in my prompt, but the AI still tends to include hits on information from the wrong version in its answers.

Going to your weapons issue, I wonder if when it comes to historical subjects, the exact meanings of terms like "halberd" also shift depending on historical context? I know from studying linguistics that words, terms, and phrases in languages are in constant historical flux.

 

[JustKneller]

What was "the right answer" in say, 2016, can be completely "the wrong answer" by 2026. Trying to program AI to deal with subjects that change so quickly probably presents a lot of problems, which are exacerbated by the fact that I as the user might be asking specifically for information about an older, outdated version of the thing, where another user is asking the same question about the current version of the thing.

I think you're well on point here. I've been using it to help plan my game runs, but I'm constantly correcting it for the IE games due to AI jumbling up classic and enhanced editions.

I also tried using it for portraits for a run, but it just keeps missing the mark. It can handle the factual details (e.g. resolution), but the more subjective details like style (e.g. doing a portrait in the style of an Icewind Dale portrait just goes over its head. And, its no surprise. It's a language algorithm so translating an image into words and then back into another image when it doesn't truly understand the components of the image is a tough task.

 

[WiseGrimwald]

I think you're well on point here. I've been using it to help plan my game runs, but I'm constantly correcting it for the IE games due to AI jumbling up classic and enhanced editions.

I also tried using it for portraits for a run, but it just keeps missing the mark. It can handle the factual details (e.g. resolution), but the more subjective details like style (e.g. doing a portrait in the style of an Icewind Dale portrait just goes over its head. And, its no surprise. It's a language algorithm so translating an image into words and then back into another image when it doesn't truly understand the components of the image is a tough task.

I have noticed that even if I get a high resolution portrait, that when imported into Baldur's Gate EE I unfortunately lose much of that resolution.

For the game the portraits are all 330 x 210 bitmaps. I am wondering if that limitation is still there in EE. That information would be helpful.

I made this quite good image of a cross between a dragon and a gnome. As can be seen, the scales can be seen quite clearly, but that is not the case when made into a bitmap.

I am aware that work still needs doing on the background, but I am happy with the foreground despite the fact that AI also put scales on some of the clothing.

As can be seen, the scales on the skin, in the game image can scarcely be seen.

 

[JustKneller]

What are you using to generate them? I've been using ChatGPT (so Dall-E by extension) because it's free. I'm trying to create some particular Icewind Dale style portraits, but it's been over a week of prompting (I only get so many renders/chats a day with the free program) and I've only just barely landed in the neighborhood of the style (though it is still rendering my subject inaccurately). I am hoping that, by the time I'm ready to actually do the run, I will have something serviceable.

I think that AI has trouble with any kind of subjective or abstract concepts. I had asked it to do a portrait, then gave it a detailed prompt to help it render the same portrait in the style of impressionism like Monet or Renoir (or Justin Sweet, to be more accurate) and all it did was make the original image blurry.