Brademante and Amyntor

Comments

• shevy123456 Member Posts: 536

  July 25

  The AI seems to be good at photorealistic images. I wonder if it can also generate a specific theme, e. g. more fantasy, less focus on photorealism. For instance, the golden/metallic armour looks quite good, even if photorealistic. The violet feather looks a bit out of place. And Minsc looks more like a slim ogre than some warrior. The parchment or pamphlet with the woman's head looks nice but the rest looks almost like a photoshopped picture. The first guy's face looks more like taken from a local pen and paper group whereas the girl looks somewhat ok. But then there is this strange ... yellow shield. Looks like from Temu or elsewhere. It's all a bit strange to me. How much can those images be influenced? Is it possible to have the images fit more towards a theme that is less photorealistic? That yellow colour is quite strange; I just noticed it is also on the parchment. Can you focus on singular entries too? For instance if you just focus on the shield part, how much can be tweaked here? Or are these images just generated once and then "forgotten" by the AI?

  0
• atcDave Member Posts: 2,962

  July 25 edited July 26

  The AI (I mostly use GPT Image 2 at OpenArt, but there are dozens if not hundreds of renderers available) can do any style you prefer. My preference is photo-realistic, I want to imagine my crew as *real* people. So I'm pointedly getting away from the BG game style. I understand many (most?) gamers feel differently about this, and that's fine. I think the beautiful thing with the renderers is you should be able to get images to YOUR liking and style preference.
  
  The violet plume is about two things, first the name "Brademante" is from Carolingian mythology, she was known as "the White Knight" and carried a white shield with a white helmet plume. I thought white seemed boring so I did something else. Initially I did not render a helmet at all. Then for a long time she wore a Helm of Charm Protection, which I noticed was depicted (in the game) with lavander colored wings on either side of the helm. So I sort of combined the two, and gave her a "helm with a lavander plume". Its different. And distinctive. I like it.
  Also the shield was something I designed specifically ("yellow triangle shield with a red rose") to be unique to her.
  
  One of the huge advances we've seen with the renderers in the last year is their ability to work from multiple source inputs. So I first work until I get a basic figure I like (Brademante is described as "short warrior woman, French, age 20, long dark auburn hair, blue eyes, fair skin, slim and athletic"). It took a few tries and a few tweaks but I finally got a face and look I liked for her. Then I used that to get her in a variety of clothes and armor, with a couple of different weapons, and the shield I'd already rendered. Now I feed it an appropriate complete model of her and describe the rest of the image. For multiple characters, I simply multiply the models I input.
  It all gives me pretty good character consistency.
  
  Most of the character/style complaints you have are on me, my taste, not on the AI per se. Although some things, like the scroll held up, is very hard for the AI. The one posted here, *I thought* was the closest to having a natural sort of hang to it. But I've never seen one yet that looks *just like* heavy parchment held up for viewing. It also doesn't know weapons very well, although that can be helped by feeding it an image of what you are wanting (bows and crossbows are comically bad).
  It also struggles if you give it multiple styles in a single render like...
  
  This is inspired by a silly exchange between Imoen and Dynaheir from NPC Project. It made me laugh, and I wondered if I could do anything with it. I first did the silly cartoony image, then fed it into a description along with models for Imoen and Dyna. I thought it came out remarkably well; BUT, it obviously shifted the whole image towards cartoony even if the style still said "Realistic photo". There is an attached image editor, I possibly could have gotten a simple image of Imoen and Dyna talking they walked along, then editing in the dialogue bubble as a seperate image. Basically forcing the issue. But I was pleased with the "flawed" image above so I used it a post.
  
  So yeah, if you look at some of my older runthrough write-ups, those were with older renderers that did not allow for image inputs. As you say, each image was unique and then "forgotten" by the AI. It sure made a bear of character of stability. You had to keep careful track of your written descriptions, then hope for the best. And it had biases like refusing to do Minsc without a beard, no matter how many ways you tried to say "clean shaven". It has been a fascinating adventure all its own. Very rapid improvements now. I feel like this current run is amazing. And it is compared to older ones, the renderers are SO MUCH more powerful now. I'm sure, this will keep improving for a while. Although of course, every new trick comes with a learning curve for the user. I'm actually DOING a lot more to construct images than I was two years ago.
  One of my other main hobbies is scale modeling, and I learned many years ago, every *better* way of doing things you learn will also take longer!
  
  BTW, @shevy123456 thank you so much for a fun and thoughtful comment!

  Post edited by atcDave on July 26

  0
• atcDave Member Posts: 2,962

  August 8

  This run of BG:SoD is now done. I think I am pleased with the result. And I like that Brad is really a mediocre sort of character, but she is smart and a good narrator. If you go back to the first post on this page is the link to the start of the SoD chapters, that will include three video renders done for this part of the story. And that involves the most recently released video renderers, that can really be terrific looking. The whole of BG1, with all of the videos made for this run is available at davesgaming.blog.
  
  
  
  
  
  

  0