Foreverlurk's AI artwork

[foreverlurk]

Hello. I’ve been lurking here forever (username checks out!), but I’ve been told by @Olo that I should share my AI stuff here. I hope it’s not against the rules or anything, I know AI can be a very sensitive topic.

My workflow is using the brand new DALL-E 3 AI from Microsoft/OpenAI as a base. Then it’s either SDXL for upscaling or face restoration, then good old GIMP if manual touch ups are required. Seems simple but Microsoft doesn’t want their AI used for porn, at all. They have TWO filters in place, so getting past them require some luck and patience.

I’m into mostly gentle stuff, so with hand helds and some clever prompts I trick the AI into drawing good sizey stuff, even some without clothes. I guess their nudity filter only applies to normal-sized women?

My DeviantArt is here :
https://for3v3rlurk.deviantart.com/

Here are a few samples.

[Giganto82]

@foreverlurk really cool images!

[tiny-ivy]

@foreverlurk these look great! I’m really surprised that AI can be used for size art this high quality. My attempts have resulted in error messages and 7 fingered hands.

Anyway my preference for AI art on this site is that it’s called out as that in the title. Which you did! So I’m happy to see you contributing this here.

[foreverlurk]

@tiny-ivy said in Foreverlurk's AI artwork:

@foreverlurk these look great! I’m really surprised that AI can be used for size art this high quality. My attempts have resulted in error messages and 7 fingered hands.

Anyway my preference for AI art on this site is that it’s called out as that in the title. Which you did! So I’m happy to see you contributing this here.

Of course I’m only showing the “good” results here, I’ve generated my shares of eldritch horrors too. Even then, none of the pics are perfect, they all have small details that are off. Still, the tech is advancing at a ridiculous pace, can’t imagine what we’ll be able to do in 5-10 years.

[littlest-lily]

I’ll be honest, I’m someone who isn’t a fan of the rise of AI art, for various personal reasons. But it warmed my heart that you were being so considerate and transparent when posting these, so thank you so much for that They do look really good!

[foreverlurk]

@littlest-lily said in Foreverlurk's AI artwork:

I’ll be honest, I’m someone who isn’t a fan of the rise of AI art, for various personal reasons. But it warmed my heart that you were being so considerate and transparent when posting these, so thank you so much for that They do look really good!

Thanks!

And I understand completely, that’s why I put front and center that this was made using AI. There are a lot of valid reasons to oppose this tech, or AI in general.

[nephilim]

WHOA. What are some of your prompts? This is exquisite.

[foreverlurk]

@nephilim it really depends, but I’ve noticed that in regards to the filters I mentionned, less is more. The key is to give the mimimum information to the AI and leave the rest up to chance. More text in your prompt can add “noise”.

My basic prompt is something like :

• “realistic [illustration|render|painting|drawing|etc.] of shrunken woman, two inches tall, tiny, shrunk, [sitting|standing|sitting|etc.] on [palm of hand|table|etc.], photorealistic”

The actual inches isn’t really important, it’s more to give another weight to the idea the subject is not its normal scale. You can then add more info about the scene, like this:

• realistic illustration of shrunken woman being examined by alchemist, six inches tall, tiny, shrunk, standing on table, in medieval alchemy lab, fantasy, digital art, photorealistic

Mind you, that took about 20 tries to get something like this. You’ll get plenty of crappy results, too.

Another thing I’m trying to do is comics… I haven’t found the perfect prompt yet but I’m getting cohesion between panels, which is a great start.

• Three-panel comic, realistic illustration showing a female wizard, in a high fantasy setting. The first panel shows the woman normal sized, she casts a shrinking spell. The second panel shows her gradually getting smaller. The third panel shows her shrinking again, even smaller.*

Be aware that Microsoft is very aggressive on censorship, like I said any negative emotions or situations cannot be prompted.

[nephilim]

@foreverlurk HAS AN ABSOLUTE FUCKING STROKE

This is incredible. Which AI system is this DALLE?

I need to try…
I’ve had Professional 3D artists & illustrators drawn my characters, who then fed their art -and my writing- into AI, creating stable heuristic models.

I need to see if I can get this to work

[nephilim]

@nephilim Here is mine.
One of my characters in a glass of water. This is actually one of my characters, and not a random seed-generated woman. I used my writing from one of my chapters to prompt this. (Premise: Heather is shrunken and put in a glass of water…)
I was able to get size difference/scale. And you can make out a tiny bit of a giant hand

[foreverlurk]

@nephilim hahahaha don’t have a stroke on me!

This is pure DALL-E, which is built by OpenAI:
https://openai.com/dall-e-3

That version will be available to paying customers of ChatGPT+. Part of its strength comes from ChatGPT4 being used to parse the weights. Part of its weakness is the censorship and filtering. Another “weakness”: unlike Stable Diffusion, you require literal datacenters to run it, and you can’t train custom models.

You can use a free version right now if you have a Microsoft Account:
https://www.bing.com/create

[nephilim]

@foreverlurk My workflow is Fotor, PromptChan, GIMP.

I have wildly uncensored stuff for my characters. Like I said, they’re stable heuristic models that I can run through various scenarios.

I want to put them through more shrinking scenarios.

See Heather in the glass above? I needed to use PromptChan for that part, where I make her very voluptuous. I have one of her naked, too; full-on pornographic nudity.

[nephilim]

@foreverlurk So, if I understand correctly, StableDiffusion lets you use your own models?

I have mine on a shelf that I can pluck off at any time to recreate scenes with from my work. It’s a vibe.

[foreverlurk]

@nephilim said in Foreverlurk's AI artwork:

@foreverlurk So, if I understand correctly, StableDiffusion lets you use your own models?

Correct, you can use and train your own. There are different technique to either add to an existing model, modify concepts inside an existing model, or training a brand new one from scratch. All with their pros and cons.

There even a few publicly available models for size content:

https://civitai.com/models/151689?modelVersionId=175451
https://civitai.com/models/92200?modelVersionId=98299

[nephilim]

@foreverlurk Thank you. Much appreciate