D&D General Using AI for character visualization

[Myrhdraak]

I have just started to play around with AI both for idea generation and creative support when writing adventures, but also for illustrations. I recently started to dig deeper into Stable Diffusion (and had to upgrade to 32G RAM on my PC to be able to do the stuff I wanted.)
I must say I find it very inspiring. I took a picture I did as a teenager in the 80's of the party druid, which was scanned in with my first scanner in the 90's and painted on my computer back then - a really impressive achievement back then .
Now 2023 I can use this picture as an input to create a totally new and very life-like picture of the druid, generating everything through AI using the original drawing as the starting point.

/Myrhdraak

 

[Blue]

I use Stable Diffusion for doing PCs, NPCs, and set pieces, like the control board of the Illithid Nautiloid spelljammer.

 

[Reynard]

It is essentially no different than googling or browsing art sites until you find an image that inspires you or reflects the character you want. I don't think anyone should feel bad for this use of image AI.

 

[dave2008]

I have just started to play around with AI both for idea generation and creative support when writing adventures, but also for illustrations. I recently started to dig deeper into Stable Diffusion (and had to upgrade to 32G RAM on my PC to be able to do the stuff I wanted.)
I must say I find it very inspiring. I took a picture I did as a teenager in the 80's of the party druid, which was scanned in with my first scanner in the 90's and painted on my computer back then - a really impressive achievement back then .
Now 2023 I can use this picture as an input to create a totally new and very life-like picture of the druid, generating everything through AI using the original drawing as the starting point.

/Myrhdraak

That looks great - thank you for sharing! I am not familiar with Stable Diffusion so I apologize if my questions don't make sense. With that out of the way, can you answer a few questions:

1. How long did it take to go from your input image to the final image?
2. What "prompts" did you use?
3. Did you have to multiple iterations or multiple attempts or both or just on go?
4. Did you use the free or XL version?

 

[Stalker0]

That looks great - thank you for sharing! I am not familiar with Stable Diffusion so I apologize if my questions don't make sense. With that out of the way, can you answer a few questions:

1. How long did it take to go from your input image to the final image?
2. What "prompts" did you use?
3. Did you have to multiple iterations or multiple attempts or both or just on go?

I can say that now that I have gotten the hang of it, I can generally bust out a reasonably nice image in 10 minutes, I usually have to go through a few iterations.

With stable diffusion, the trick is the eyes. Even with the later updates, its very bad at eyes. But what you do is use their in-image technology to highlight just the eyes, and then run like 40 versions of just eyes until you get ones you like. That's probably the longest part of most images 've done.

In terms of multiple attempts, you can set a batch size of 8, and as many batches as you want (just depends on your computer hardware, a lot of memory and a strong graphics card are big factors). So once I feel like my prompt is giving me reasonable stuff, I will run like 40 images at a time, and then pick from teh litter the ones I like.

Also pro tip, chat gpt is pretty decent at generating stable diffusion prompts. So a lot of time I'll start with that, and just go to work.

 

[dave2008]

I can say that now that I have gotten the hang of it, I can generally bust out a reasonably nice image in 10 minutes, I usually have to go through a few iterations.

With stable diffusion, the trick is the eyes. Even with the later updates, its very bad at eyes. But what you do is use their in-image technology to highlight just the eyes, and then run like 40 versions of just eyes until you get ones you like. That's probably the longest part of most images 've done.

In terms of multiple attempts, you can set a batch size of 8, and as many batches as you want (just depends on your computer hardware, a lot of memory and a strong graphics card are big factors). So once I feel like my prompt is giving me reasonable stuff, I will run like 40 images at a time, and then pick from teh litter the ones I like.

Also pro tip, chat gpt is pretty decent at generating stable diffusion prompts. So a lot of time I'll start with that, and just go to work.

I looked at the free online version and I didn't see a way to start with an image (it was just prompt based). I have a lot of old art work that I have done and would love to run it through this. Am I missing something or do I need to purchase it to use all the functions.

 

[Stalker0]

I looked at the free online version and I didn't see a way to start with an image (it was just prompt based). I have a lot of old art work that I have done and would love to run it through this. Am I missing something or do I need to purchase it to use all the functions.

So I am using the automatic1111 interface which is a GUI that helps you run the system. I highly recommend installing stable diffusion on your computer.

how to install stable diffusion on windows - Search Videos

www.bing.com

Here is a video to help. Now the first install can be daunting, just take it step by step nice and slow, but once you set it up you can use it as much as you want with no cost!

 

[Myrhdraak]

That looks great - thank you for sharing! I am not familiar with Stable Diffusion so I apologize if my questions don't make sense. With that out of the way, can you answer a few questions:

1. How long did it take to go from your input image to the final image?
2. What "prompts" did you use?
3. Did you have to multiple iterations or multiple attempts or both or just on go?
4. Did you use the free or XL version?

Well I am no expert yet and I do not have a very expensive PC so each rendering take a while so it took me most part of the evening to get it where I wanted. I would say 3 hours maybe.
I used the free Stable Diffusion version and I found these really good instructions from another swedish guy, which I would recommend (you can pick up som dad jokes as well):
https://www.youtube.com/@sebastiankamph/videos
I had to do a lot of iterations (both txttoimg and imgtoimg) and I used photoshop in between sometimes just to sharpen the image or cut and paste object into the picture in order to steer the AI in the right direction (the tree trunk on the right I had to "inspire" the AI to do for example. But you could have used GIMP or some other program that is free for those basic things as well.)
As for prompt, it was something like "A 30th year old male druid with a blond hair and beard, intense eyes, standing in a subtropical forest". However I used Sebastian Kamph's Stable Diffusion styles - "Negative" and "Digital Painting" which adds a lot of control words to get it right. See more in his videos.
Hope that helps.

/Myrhdraak

 

[cbwjm]

My friend has done many prompts via midjourney which come out looking epic. Some are great for scenes, others are great for character concepts. AI art is perfect for this stuff, assuming you can get the right prompt.

 

[dave2008]

Well I am no expert yet and I do not have a very expensive PC so each rendering take a while so it took me most part of the evening to get it where I wanted. I would say 3 hours maybe.
I used the free Stable Diffusion version and I found these really good instructions from another swedish guy, which I would recommend (you can pick up som dad jokes as well):
https://www.youtube.com/@sebastiankamph/videos
I had to do a lot of iterations (both txttoimg and imgtoimg) and I used photoshop in between sometimes just to sharpen the image or cut and paste object into the picture in order to steer the AI in the right direction (the tree trunk on the right I had to "inspire" the AI to do for example. But you could have used GIMP or some other program that is free for those basic things as well.)
As for prompt, it was something like "A 30th year old male druid with a blond hair and beard, intense eyes, standing in a subtropical forest". However I used Sebastian Kamph's Stable Diffusion styles - "Negative" and "Digital Painting" which adds a lot of control words to get it right. See more in his videos.
Hope that helps.

/Myrhdraak

So I am using the automatic1111 interface which is a GUI that helps you run the system. I highly recommend installing stable diffusion on your computer.

how to install stable diffusion on windows - Search Videos

www.bing.com

Here is a video to help. Now the first install can be daunting, just take it step by step nice and slow, but once you set it up you can use it as much as you want with no cost!

Thank you both