Tutorial: Pixel Metamorphosis with GPT Image 2

I said it before, but I think one of the best things about GPT Image 2 is its ability to transform things and images.

Think about New York City, for example. There are many photographs or movies about it, and I think most people will know the iconic "look" of the city.
But how about we twist and mix things up a bit? Let's create a pixel art transformation of NYC.

Prompt: "an image of New York in 16 bit pixel art style."

Not bad, right? But let us add some depth:

Or how about a night time view?

Prompt: "an image of new york at night in two dimensional 16 bit pixel art style."

Or maybe we could move the city into the future, and add a touch of cyberpunk?

Prompt: "an image of new york in two dimensional 16 bit cyberpunk pixel art style."

I think it is remarkable that GPT Image 2 is able to transform the "look" of New York into completely new settings, context, concepts... and adheres to the prompts that we come up with.
And, even though the image, or the "view", is totally changed, most viewers would still immediately recognize that it is NYC!

Let's move on. How about other cities?

Prompt: "an image of london at night in two dimensional 16 bit pixel art style."

Prompt: "an image of hamburg at night in two dimensional 16 bit pixel art style."

Prompt: "an image of los angeles at night in two dimensional 16 bit pixel art style."

And maybe we can change perspectives now?

Prompt: "an image of los angeles at sunset in isometric 16 bit pixel art style."

And now that we did this... as it is pixel art, maybe we can have a little "fun", and transform the image into a video game setting?

Jump and Run / Platformer Game

Chess Game (Showing a match between LA and NYC!)

Isometric Puzzle Game

There might not be an immediate "use case" for this tutorial. But it tries to show that GPT Image 2 transforms images well, and even adheres to very unusual prompts and requests. And can even change or add dimensions or put everything into a whole new setting.

Tutorial: How to let GPT Image 2 do the creative work

Tutorial: How to let GPT Image 2 do the creative work and come up with several characters and objects that fit to each other at once

There are two things GPT Image 2 is very good at. They are like an inversion of each other.
Following prompts very strictly, and in a very specific way.
And the opposite thing, to be able to do brainstorming. To "fill in the gaps" of a prompt. To take over a lot of the creative workload of a designer.

The following examples are about pixel art designs, but could be used for any type of design work really.
What is the objective of the tutorial? Well, often an art designer needs a number of elements, characters, objects, designs, tiles, backgrounds...
And when using AI, usually the designer would need to come up with neat prompts for that.
Say, the designer wants to create characters for an RPG game, then that designer would need to conceive all these characters in their mind. This can take a lot of time and brain energy. Hours, maybe even days!
Same when coming up with biomes, or buildings for a world building game, rooms of a space station, equipment for an RPG hero...

But, in my opinion, GPT Image 2 is very good at taking over this brainstorming job, too!

Let us a look at some examples:

Prompt: a 3 x 2 grid with 16 bit pixel art designs of the party in an cyborg themed rpg

Prompt: a 3 x 2 grid with 16 bit pixel art designs of the vehicles in a futuristic war game.

Prompt: a 3 x 2 grid with 16 bit pixel art designs of the tiles in the biomes isometric exploration game

Prompt: a 3 x 2 grid with 16 bit pixel art designs of vehicles in a fantasy racing game

Prompt: a 3 x 3 grid with 16 bit style pixel art designs of a the rooms of a futuristic space ship

Let me explain what is going on:

GPT Image 2 follows the prompt, and comes up with multiple designs for the "job".
The designer needs to take less focus on coming up with everything on their own, and could even use the AI as an inspiration for own designs.
Most importantly, the various objects and individual designs *align* and fit to each other, and are part of the whole concept.

Of course, the designer would not need to use all of these. They can generate multiple "grid images", by re-using the same prompt. And then use these from each grid that work best, discard the rest.

These images might be quite small at times; but with AI, it's of course possible to cut them out and resize them to full size.
In the same way, the background of some of these objects could be removed easily.

These were still some quite basic designs. I am sure even more complex and better things are possible, when using this technique!

Bonus examples:

Tutorial: Staying within the Grid - Creating pixel artwork characters with GPT Image 2

Hello Friends,
And welcome to another tutorial for GPT Image 2.
While most of my other tutorials were very versatile and could be used for most purposes, this one is quite specific.
Specifically, it is about video game artworks, or pixel artworks in general.
When trying to come up with art like this, it is often important that the characters, creatures, units, buildings, are approximately the same size, don't "bleed" into the terrain, don't clash with other objects, and so on.

GPT Image 2 is well suited for that, and in the tutorial I will show you plenty of examples and use cases.
But it wouldn't be a tutorial without an explanation of the workflow, and more details, right?
So, the "key" for creating arts like this, and also for creating your own, custom prompts, is the following:

-You need to be very specific and state exactly what the art should *be* like, which ideas or "rules" it should follow.
For example, I first tried to use prompts that specified a grid playfield and the theme of the art. But I realized, this was not enough. The characters began to cross the boundary.
What I did add was a line that explained to GPT Image 2 that this should not happen in the finished artwork.

-When doing the artwork with top down, or horizontal 2D graphics, a new problem happened: the art became very chaotic, characters had different sizes...
Therefore, I also specified that the entire art should follow, and be placed along "grid" or "tile logic".

And this fixed the thing!

-With some graphics, for example isometric designs, it is not desirable to avoid "bleeding", as the characters need to be larger than the tile they stand on, usually. To compensate for this, I specified that the character art should be close to the same size for the major characters.

So what is the use for this?
These artworks could be used as design drafts, or, if the background gets removed, the characters could be utilized for games straight away.

Examples:

Prompt: create a cyberspace themed tile based isometric retro game with 16 bit style pixel art characters. all characters should approximately have the same size

Prompt: create a tile based top down 2d strategy game field with 16 bit style alien themed pixel art characters. the character artwork should stay within the boundaries of its respective tile.

Prompt: create a worldbuilding game style hexagonal playfield with 16 bit style pixel art buildings. the building artwork needs to stay within the boundaries of the respective hexagon tile.

Prompt: create a labyrinth game with 16 bit style pixel art characters. walls and characters should be placed according to tile logic. the theme is libraries.

Prompt: create a platformer game with 16 bit style pixel art characters. walls and characters should be placed according to tile logic. the theme is ice and crystals.

How to get GPT image 2 to tell a story from start to finish by (more or less) using a single prompt

Having fun in the 4th dimension (time) or: how to get GPT Image 2 to create a sequence of images that tell a story from start to finish by (more or less) using a single prompt

Hi Friends,
This was one of the moments where I was really, really blown away by the capabilities of AI, or GPT Image 2, to be more specific.
Generating "still" images is all fun, but sometimes we need to create a series of images that have a progression, for example for telling a story from start to finish. we need to add the "fourth dimension" to our images - time.

Usually, this is done by the user. you need to think up your story, flesh it out in detail, and then you try to truncate each part of the story into small segments, that could fit into one single prompt. and then you try to come up with the right prompt for each.
For example: one prompt for creating an image of a character that walks to a spaceship, one prompt for lift-off... and so on.

But wouldn't it be nice if we could delegate this task to AI? And find a prompt that generates sequential images, without needing to change the prompt too much?

And here is that one.

For this example, I want to create a futuristic pixel art short story. So I use this prompt:

i am working on a short story in pixel art style. the story should have 7 pages. it should tell a sci fi story beginning on earth, explore strange and alien planets, and finally end on earth again.
please generate page 1 out of 7 now.

Now all i need to do is to increment the number for each new image. by changing the last line to "please generate page 2 out of 7 now.", "please generate page 3 out of 7 now."... you get the point.

If somehow, an image that get's created does not fit too well, we can re-use the same prompt, until we are happy with it.

I am using Leonardo.Ai for this, and GPT Image 2 is one of the models they have available.

so let us look at a test run for this:

Isn't this stunning?
All these images were created by repeating the prompt, and just increasing the number each time. I did not change anything else with the prompt, nor did I add things, or more specifications.
You could say some elements are slightly off and misalign a bit. Personally, I think the art style could be even more coherent. But this is just a test run, and with some tweaking, this problem could be solved, too.

And, once again, "there is more out there", and this technique can be enhanced and researched even more!

Tutorial: How to use GPT Image 2 in the same way as ChatGPT - but with a visual twist

GPT Image 2 is out for a while now, and has been blowing everyone's mind, including mine.
It is more responsive and "understanding" to prompts, tasks, specific demands than any other AI image generator I tried so far.
It's so smart and advanced that you can actually use it... or rather, talk to, in a very similar way as you talk to ChatGPT - and use it for most of the tasks that you set for ChatGPT, too! But with a visual twist.

And that is the big, big game changer. Because, so far, the various AI models were separated by their mode. We had large language models like ChatGPT or Gemini, AI image generators, AI sound generators, AI video...
But GPT Image 2 more or less merges the power of ChatGPT and image generators into one.

So, you can talk to the AI, just like you would talk with ChatGPT.
The only difference is that this time, the AI does not respond with text, words, chatting, or at least not directly. It responds in a visual way.

So let us start with some examples:

Prompt: Rank the 5 tastiest italian dishes. Give a reason for each.

Prompt: Create a pixel art design, in which nikola tesla explains his invention of alternating current.

Prompt: Create an info graphic explaining the australian emu war. The graphic should look like it is actually from the 1930s.

Create an info graphic that explains the differences between a trebuchet and a catapult. Make it look like it's from the medieval era.

Explain what a labyrinth is. The letters should be arranged like a labyrinth or maze themselves.

Tell me a good italian spaghetti recipe with which i can impress my guest. make the recipe look like it was written on a medieval scroll.

Create a pixel art design that shows a space station control room. There should be a screen, and the screen should show 5 interesting facts that people rarely know about english grammar.

What is the use for this method of talking to GPT Image 2?

Well, at the most basic level, there are at least 3 potential use cases:

1: "spicing" up your text output. Want to create a promo text for your new steampunk metroidvania game? then let it create the text *in the visual style* of the game.

2: creating very specific artworks and visuals, with long and fancy texts, sentences, passages...

3: creating images where the visual arrangement of texts is actually vital to the image. crossword puzzle designs, labyrinth structures made up of sentences...

These are just some basic examples. I think there are still boundless other uses possible... there is still a lot of research that can be done!

Tutorial: A little hack to use GPT Image 2 for brainstorming and creative drafts

I think a lot of people know this problem: generating ai art can make things easier and have amazing results, but actually finding your way to these "good" results can take a lot of time. and eat up a lot of energy!

i'll give an example: i wanted to create a kind of futuristic cyborg DJ, but brainstorming how they are gonna look like, and then (essentially) doing trial and error with the ai generator... well, I wasted hours on this.

Sometimes one needs to spent a lot of time *thinking* about the art one wants, and how to phrase it correctly to the ai generator, before things finally work out.

So, by accident I stumbled upon a little hack with GPT Image 2, that bypasses this bottleneck and gets one quickly to these sweet results.

I am certain I am not the only one who "discovered" this, but I want to write about it anyway.

The idea is: you can give a *general* idea about the artwork you want, and tell GPT Image to give you a set of tiles with the artwork.
a grid, for example of 3x3 tiles or 3x4. that would give you 9 or 12 different images.

and then gpt image 2 does the whole brainstorming session itself, you don't need to worry about it anymore. it generates a number of possible artworks, and you can simple download the image and cut out the one you want to use. or generate a new set of tiles again, until you are happy.

how is this different from just generating 9 "full" images in a row, by repeating the same prompt?

well, as you can see by the example images, gpt image 2 is *much* more creative and diverse with the generated images this way. if you just prompt it 9 times with the same prompt, it is likely to generate very similar images. but not with this method!

obviously, this is suited for specific artworks, like icons, pixel art, game characters, and such... that do not exactly need a high resolution.
but it could also work for high resolution images: generate characters this way, cut out the one you want to use, then run it through gpt image 2 with image reference, and it will blow your character up to "life size!".

now let us look at some of the possible uses for this hack.
just like with the other tutorials, i will go with a "retro pixel art" theme, but for your own images, you would not need to do this.

prompt: create a 3x3 grid of 16 bit style futuristic pixel art characters.

so we have a selection of 9 images now. if i like the top right one and want to use it for my game, i can cut it out and discard the rest:

we can also be very specific about the theme. for example an rgp taking place in an egyptian setting.

prompt: create a 3x3 grid of 16 bit style pixel art characters for a rpg with an egypt theme.

or if we want to have variations of a specific character.

prompt: create a 3x3 grid of 16 bit style pixel art characters for a futuristic game. cool space rogues with green hair.

the user could then, for example, decided to use the top left one for their game.

but it does not even have to be humans or creatures!
how about some space ships?

prompt: create a 3x3 grid of 16 bit style pixel art space ships for a futuristic game.

or buildings for a world building game?

prompt: create a 3x3 grid of 16 bit style pixel art buildings for a world building game. isometric tiles with futuristic buildings.

the possibilities are, once again, endless.

bonus images: