In for a Penny
Added 2023-06-29 01:05:13 +0000 UTCHey guys! I'm still working on the scenarios, I'm really liking the result. The idea of using modular parts to build the buildings made the work much simpler and with cozy aspect. I now need to furnish the rooms and apply the textures and this step will be done.
In the meantime, I made one more attempt at using AI, this time I applied a background to Penny's images, apparently the AI understands images better this way than against a gray background, which is how the images were before. I also added more detail to the description of the images when training the AI, and the result was interesting.
The AI understands details like hair color, freckles, breast size, and hip size, but there is still variation in proportion. When I try to generate a realistic image, the face shape looks very similar to Penny's in 3D, however when I try a Pixar-style image, the face comes out with a generic shape. I tried a prompt to generate a middle ground, and here are the images I was able to generate.
I believe that it would be possible to generate consistent results as long as I fed the AI with a sufficient amount of images, however, it would take a lot of time and work to accomplish, which is far from being a priority. But I want to continue spending a couple of hours a week trying to improve the results, and I'll post it here so you can see the progression.
Thank you very much. See you later.
Comments
Hello, thanks! I'm glad you like my style of characters, it took a lot of work to achieve. Yes, with AI I can create details for the textures, and I can keep everything similar and uniform in style. But the main reason I want to feed my characters to an AI is to create static images like wall pictures and collectibles, for instance, Brendon will find an old magazine with pictures of one of the characters. Pre-rendered images look best as still images, but achieving good results takes time and skill. The AI can take a not-so-nice render and make it look perfect, as long as it understands the proportions and style characteristics. But while I don't provide enough data, I need to try to anchor in other styles, like Pixar. I fear that perhaps I have lost the sensitivity to be disturbed by AI-generated images after seeing several attempts go very wrong. I'm a little worried, maybe I should wait to get really consistent results before sharing these images with you. I'll keep an eye out for feedback.
MaliXe
2023-06-29 11:43:27 +0000 UTCHello, Thanks for the news. Great if you've managed to make things easier for yourself with modular environments. It's true that it seems simpler to do this option than to create an "open world". Can't wait to see the result when the time comes! Keep working on the project, little by little. For the Ia, I don't know if it's because I'm used to your graphic style (which I like a lot) but it always disturbs me. En core plus dans le style Pixar XD. So if I've understood correctly, with IA you can then program it with your own graphic style to facilitate certain tasks, notably texturing? Surely it can't be easy. Thanks again for the news! Kind regards
Archy_TaxiWhy
2023-06-29 06:44:19 +0000 UTCYou're welcome! 😁
MaliXe
2023-06-29 01:18:17 +0000 UTCThanks for the update!
Mike C
2023-06-29 01:08:06 +0000 UTC
