AI General Sharing Thread

Decided to create one final golem. The Manure Golem. If you are as old as me you might remember this thing from one of Spoony old Counter Monkey videos. Where these things were responsible for creating the infamous "Age of Manure".


Decided to include this one as a reference.

Gemini is inconsistent with its filters. Says something and if you try again and get lucky, does the opposite of what it said.
Pretty fun to test its limits.

Note that gemini is not generating any images, it is a completely different AI, and gemini has no idea what image it generates looks like. If it generates a turd and you say it generated a princess, it will believe you. But when you als anything about the picture, it answers correctly (mostly), because that is the first time it analyzes the result.

It is funny that AI is called intelligence, because it is dumb as... uhm.... as something stupid. And seriously, any 6th grader can do better math.

Gemini displays zero intelligence. It's a randomised robot with limits and pre-established answers in some cases.

Not sure about the robot knowing or not what the elements on an image are. It does seem to censor what the humans programmed it to.

When I refer to Gemini, I refer to whatever it coordinates with to generate images. Which I believe is Nano Banano or something like that.

And AI is pretty good at analyzing and describing an image. I had it analyze and describe my tattoo and it was pretty spot on.

However it does it, it usually nails what I am after with the description I provide first try. I would usually need a few tries with other generators, but for whatever reason, this one gets my vision immidiatly in most cases.

Maybe it is the thing it coordinates with that is really good.

@Darkmoon9 these monsters really well
possible to creat types of plant monsters and less humanoid???

Sure, though it was a little bit of a challenge.

For those who are interested, have an in depth chat with Gemini about how it generates images or stories.

You can ask it a somewhat general question and then drill down on certain aspects. My chats have been pretty interesting so far in trying to peer behind the curtain so to speak.

I recently messed around with Google AI Studio, and using its app builder, I made a sort of MV/MZ assistant AI. It works surprisingly well, featuring face/bust/sv battler generators, plugin creation/edit/bug fixing, as well as eventing.

Anyway, here's the link if you wish to try it out

link

We love you, Harold! (It's the RPG Maker Jam!)

In the spirit of educating oneself on a topic so you can speak intelligently on it, I asked Google Gemini the following question:

"I know that when you create an image you access Nano Banana. But how are images actually created. Does AI "steal" from human artists to piece together a new image? I don't think it does. If not, exactly how does that work? If I ask for an image of a "troll", how does AI know what a "troll" is and how does it create a new image of a "troll"?"

Here is the response:

This is a great question. You've hit on the exact distinction that separates "AI generation" from a "collage."
The short answer is: No, the AI does not steal pieces of art and stitch them together. If it did, you would eventually find the "seams" or recognize a specific eye from a famous painting or a particular arm from a stock photo. Instead, it works more like a human brain remembering the concept of an object rather than a specific photo of it.
Here is exactly how the AI creates that "troll" from scratch.

---

1. How it learns what a "Troll" is​

During training, the AI is shown billions of images paired with captions (e.g., an image of a bridge with the text "a stone bridge with a green troll underneath").


It doesn't "save" the image. Instead, it uses a process called Pattern Learning. It looks at thousands of images labeled "troll" and notices statistical similarities:

• Geometry: Trolls often have large noses, hunched backs, and mossy skin.
• Environment: They are frequently associated with bridges, caves, or forests.
• Style: They are often rendered in "fantasy" or "concept art" styles.

The AI turns these observations into mathematical weights (numbers). It effectively learns the "recipe" for a troll. By the time it's finished training, the original images are gone—all that's left is a giant mathematical map called the Latent Space.

---

2. The "Static" Method: Diffusion​

When you type "a troll," the AI doesn't go to a folder of troll images. It starts with pure random noise—literally a screen of "TV static."


The creation process is called Reverse Diffusion. It works like this:

1. The Noisy Canvas: The AI starts with a block of random pixels.
   
   
   
2. The "Denoising" Step: The AI asks itself, "If I shift these random pixels just a tiny bit, which direction makes this look 1% more like a 'troll'?"
3. Iteration: It repeats this thousands of times. It slowly pulls "troll-like" patterns out of the chaos, like a sculptor seeing a shape inside a block of marble and chipping away the parts that don't belong.
   
   

$$x_{t-1} = \mu_\theta(x_t, t) + \sigma_t \epsilon$$

---

3. Why it isn't "Stealing"​

Because the AI is predicting pixels based on probability rather than copying them, the resulting troll is entirely new. It has never existed before.

• Human Artist: Sees a thousand trees, remembers "trees have bark and leaves," and draws a new tree.
• AI: Analyzes a billion pixels of trees, calculates the "mathematical essence" of a tree, and "denoises" a new tree from static.

While the AI was influenced by human-made art (which is why there are huge ethical debates about whether artists should be compensated for that influence), it is not "copy-pasting." It's more like a student who has looked at every painting in the world and is now trying to paint something in that same spirit.

---

This is consistent with other sources that I have read on this specific issue and I find it helpful to better understand the process rather than blindly believe generic, negative statements like AI just steals artists work. The reality is much more nuanced than that.

A human artist certainly takes inspiration and even subconsciously copies, to some degree, other content that they have seen, drawing from the totality of their experiences. But we wouldn't say that someone how grew up learning from Bob Ross is just "copying and stealing" from Bob Ross.

If you remain unbiased, you can have some very interesting discussions with Gemini and really drill down into things. And it is sort of fun when you point out a contradiction in what it produces and it has to correct itself. Fortunately, in my experience, it actually acknowledges those contradictions better than many humans.

Thats intresting, so it works more like a human artist than I thought. Explains why I never noticed a piece of a human made artwork in a piece generated by ai.

Also, here are some more images of Dahlia as a warlock, rogue and cleric respectively. I think I have managed to make the hair more consistent.

Edit. Forgot to attach images. The 4th one is an alchemist.

Someone asked me a question unrelated to AI but I think it might fit if we ask it about it too.

If someone wasn't going to pay any artist or coder or whatever to do anything, regardless of if AI is here or not. And they start using Ai to make those things. Is that bad? Under no circumstances would they have paid a human to do it regardless.

Kind of reminds me of my old argument of why I used to pirate xD I never intended to buy the things I pirate anyway so they didnt lose any money (on me specifically) the only exception was when I pirated something to see it faster or if i liked something. I ended up also buying it, which is good for them :p

I know what you mean. Thing is, I could never afford to commission for all art in a game. Even with my first game, Dextera Mortis I only have commission art for some key characters, mostly party members. And there is a good chance some of the commission artwork might be ai anyway, which defeats the entire point of commissioning art. Most portraits are created with the built in Character Generator. And even this set me back like a 1000 dollars or more. Which is a big ask for a one man project I released for dree.

I could only afford to spent so much money on one project anyway. For Chronicles of Avineria the question is more between Character Generator face portraits or ones made by ai. Both of which doesn't cost me anything.

Truth of the matter is that I can't really afford to commission art to begin with. So no sale is truly lost.

The thing is, AI is indeed trained on so many images, but the images it was trained on contain copyrighted materials, copyrighted characters, copyrighted designs.

A simple example: If you ask AI to create an image of a themepark, chances are high you might see some disney characters. Even if you never ask for disneyworld. The thing is that so many images named with tags like themepark have disney characters in them.

If you ask for a comic style, you might get characters and drawing while that look a lot like Stan Lee's work or DC characters. You might even see the superman logo appear when you ask for a super hero. In fact, it is pretty likely (depending on which AI you use). Hardly ever batman or spiderman, always superman. :p

Those disney and comic characters we know, but what if a brand or character slips in there that we don't know? Then we think we have an original character in there, while it is just an unauthorized use of copyrighted content.

That is why AI is still on debate. According to European law, we are allowed using the output, but that doesn't mean the AI is not ripping off existing art. It is.

Kinda simple really, just don't use the material containing copyrighted characters officially published work, but it is fine for private use.

In fact, you can use this to create images of x character as a cleric or something like that. I have used images like that for the player character in games like Pathfinder Wrath of the Righteous. That is the cool thing woth games that allows ypu to upload any image for the player character.

I see you're in Sweden, and according to EU law, you are even allowed to use it for commercial use.

But the point of view from te artists is different. Don't get me wrong, I am very much pro AI. But I just do see how it actually does hurt artists.

It is a knife that cuts on two sides, but it s a change that is happening and we couldn't stop it if we tried. We can either get on that train now, or miss it completely.

I am absolutely not saying that using AI is bad in any way. This is pretty much the only advantage we have on American artists, so take it and take it now. (Since in USA you risk being sued.)

It might seem unfair, but in return Americans can use ANY content that enters te public domain 20 years (!!!) before we can. They can use lord of the rings since 2024, while we have to wait till 2044.

So enjoy your AI creations, and take the advantage of being a EU citizen in this particular case.

Just one thing, before you use it publicly, or even more important if you use it commercially: Actually do read the terms/conditions of the AI you use. I made that mistake with Udio... I thought I could make music and post it on my channel, turned out they claim all the rights, even though I'm EU. (And they were such a-holes about it, too.)

The way I see it is that the way ai will affect artists, as well, as other jobs that might become redundant is that it is similar to how the invention of the printing press affected those who's jobs were copying books by hand. Some jobs will always be affected by new technology, but there would still be people who needs to run the printing presses so to speak.

Thing is, more than just artists will have their jobs replaced by machines. But I will never argue we should stop developing new technology. Rather I consider the societal structure where people need jobs to survive to be the problem. We are slowly approaching a future where almost anything is automated and human labor not being required for most tasks.

I think people should get used to the idea of doing things because they want to so it, not because they need to. If you like to draw, then draw. Or like in my game, I have never got a dime from creating games and I have paid alot to create my vision. But I consider the point of this endeavour is to express my ideas. Not neccesary to get rich, it would be a bonus, but it is not ultimately why I do it.

And thanks, I should really read the terms of service of Gemini, the only problem is that such agreements are usually really long and hard to understand for people outside of studying law.

I have been exploring a wide variety of AI image generators with mixed and interesting results. I am only using the free versions, which means that some results are of lower quality or have watermarks.

Here is a list of most of the generators I have been exploring, using their identifiable names or URLs:

• Gemini
• Grok
• ChatGPT
• Stable Diffusion
• Canva.com
• Raphael.app
• easemate.ai
• perchance.org

They each have different levels of safety filters and have different quality outputs. Some generators have super restrictive safety filters while others seem to have almost no filter at all. And some generators don't follow the prompts as well as others and they have issues with writing. And some have problems with interpreting ages as well.

One of the things I like specifically about the perchance.org generator is that it has a huge set of Art Styles to choose from, some of which are very high quality and pretty cool. But its downside is that many of the styles have obvious, low quality mistakes and the interpretation of the prompts is no where near as good as other generators.

I decided to see what each generator would do with a generic prompt for a sample book cover for a Zombie Apocalypse trilogy I want to focus on finishing this year now that I'm retired. Since the main character of the book is a 13 year old boy, I tried to keep the prompt very generic to avoid problems with safety filters that don't want to show minors in certain situations. If you're curious, ask Gemini what its safety filters are for showing minors in a horror setting.

Here is the prompt that I used for all generators and curiously, Gemini is the only one where the prompt triggered an immediate safety filter. I had to ask Gemini to create an image with the concept that also complied with the safety filters.

Prompt: "I need cover art for a book about a Zombie Apocalypse where the Main Character is a 13 year old boy. Create a book cover image, without a title, that captures the Zombie Apocalypse setting with a young, 13 year old boy protagonist."

Here are the various results by generator under Spoilers. And I didn't try to refine or duplicate any image generation, but I did try various Art Styles with the perchance.org generator.









Anyway, I thought this experiment was very interesting.