EASY GUIDE TO CREATING AMAZING A.I. IMAGES USING MAGE.SPACE

As any regular readers of my blog will know, I’ve formed something of a love/hate relationship with artificial intelligence over the past eight months or so. Some days, I lose myself for hours in creating AI-generated art, while on others the spectre of what it means for the future terrifies me. But, still, I’ve created a lot of AI images, and I think I’m getting pretty good at it. Well, I get quite a few people asking how I make them anyway. So, here’s a quick guide to my preferred AI image platform, mage.space.

First up, I must stress, I have no affiliate agreement with mage.space whatsoever - I make no money for referring people. This is simply an honest guide to what I’ve found to be the best and most flexible AI image-generation platform. Also, there is a free option, but don’t bother wasting time with it because it is very limited. Instead, just go straight for the premium account - it’s $15 per month for unlimited use. I mean, really, that’s equivalent to a couple of decent beers or coffees these days.

What is mage.space?

In a nutshell, mage.space is a platform that brings together a growing range of fine-tuned AI models, based on various AI programs including Stable Diffusion and Midjourney, and makes them accessible through a very intuitive user interface. But, don’t be fooled, this isn’t dumbing things down. Yes, ‘easy mode’ cheats and simple settings, but as you get more confident, you can take control to a very granular level.

You can create images entirely using text prompts, or through a mixture of text and image prompts (my preferred method), and all iterations are automatically saved for you. There’s also a friendly and vibrant Discord community, and you’re given the option of creating your images in private or sharing them publicly.

Getting started with mage.space

So, you’ve signed up for a pro account and are ready to get creative! Here, I’ll share a few tips to get you off to a flying start. But, before I do, here’s something to always keep in mind: the process of actually creating an image might be as easy as clicking a button, and take just minutes, but, to get the image you actually want - the image in your mind - takes patience… a lot of patience. Everything is about iterative evolution.

For this example, I’ll use a combination of text prompts and an image prompt - I find it gives the best results (or you at least get better results quicker). Luckily, because I’ve been a passionate photographer and artist for decades, I also have a ready supply of unique, original images that I own the rights to. But, if you don’t, you can always find some free stock images or even go out and snap new photos - the base prompt images don’t have to be great, but just enough to give a guide.

Choosing a good prompt image

Righty, so I want to make an image of a beautiful young redhead woman having fun at a waterpark (which seems apt for the hot summer we’re having here in Malta).

So, I’ll start by finding a picture I took a few years back of one of my friends, the beautiful British model Lottii Rose. Now, I’ve got hundreds of pictures of Lottii floating around my hard drives, but to get the best result it is important to choose carefully. I need to think about what I want the final image to look like.

So, for starters, it would be best to have her wearing a swimsuit or bikini but also to be in a pose that is at least similar to the one I envisage for the final image. It doesn’t have to be the same, but the nearer it is, the easier AI will find it.

At this point, keep in mind that, so far at least, the AI models find it easier to deal with figures who are standing, sitting, or kneeling. Basically, anything where they are upright, rather than horizontal. It is possible to create lying or reclining figures, but it takes more time and patience to get them right.

Then there’s the background. Now, it is possible to mask the background or foreground in mage.space, and deal with them separately, but, you can also make life easier by selecting a specific prompt image. For example, if I want the girl in the final picture to be in a swimming pool, then using a prompt image where Lottii is in the sea or water will help give AI a good cue. Likewise, if I want the girl in the final image to be standing near a waterslide, then a source image with a relatively cluttered background will help. The visual cues don’t have to be precise - just a rough approximation to ‘inspire’ the AI.

So, this is the picture I’m going to use as an image prompt. She’s in an easy pose, with the right visual cues around her to help AI get on the right track for my envisaged final image; a girl wearing a swimsuit, having fun at a water park.

Remember, I’m not attempting to make a final image that looks like the original prompt image, or even the same person, this is just a guide for the feel, theme, and structure of the image. The same principle applies whatever the subject, whether it’s a person, landscape, vehicle, spaceship, animal, or anything.

Writing good text prompts

This is where things can get confusing. If you look at the examples provided on mage.space, or by much of the community, text prompts tend to be very long and detailed. However, I find it much easier and more effective to start simple. Very simple.

You see, the thing to remember here is that AI image generation, at least in my opinion, is not about writing a highly specific prompt that will then lead to a highly specific outcome. It is about working with the AI to nudge and iterate, to encourage it to share your vision and inspiration. Even AI experts acknowledge that, despite humans creating AI and helping it develop, we don’t actually know how it works at the most fundamental level. We simply put information in, train it, and it crunches numbers and spits things out - it doesn’t follow lines of set code.

So, start really simple, and gradually iterate. When the AI starts generating results that are in-line with your vision, then double-down on those and nudge it in that direction. So, in this example, I’ll start with this very simple text prompt:

“Gorgeous young woman, wearing one-piece swimsuit. Dripping wet. At waterpark. Kodachrome-style photograph.”

Choosing the best AI model

Now we have a good prompt image and prompt text combo to start. Next, we need to choose which AI model to use. This is where you might start to feel overwhelmed. But, don’t worry, just be patient. At the time of writing, there were over one hundred base models available at mage.space - and you can even combine base models with specialised detail models. Luckily, there is a useful menu that allows you to sort the models based on the following categories; Art, Photorealism, Anime, Celebrity, Exclusive, NSFW, Inpainting, and 3D-Art.

For this particular project, we obviously want to use a base model optimised for Photorealism. However, even within this category there are dozens of options. So, the only thing you can really do is take the time to experiment. After a while, you’ll get a feel for which base models will be most suited to your desired outcome. Have fun, don’t rush, and you’ll soon have some favourites.

For a girl at the waterpark, we’ll use CyberRealistic - it’s one of my favourite, though I also like Consistent Factor, Deliberate V2, and Hassan.

We’ll keep things simple for this project, so we’ll just work with a base layer. As you get more confident, you can combine the base layer with additional specialised layers, but that isn’t needed here.

Prompt strength and other settings

Now, we’ll need to run through the initial settings. It might seem frustrating, but getting these right will make things much easier later. There is an ‘Easy Mode’ available for most models, which will pre-set things, but I don’t recommend it. The best way to get a good grasp of the process is to tweak settings yourself.

First up, set the prompt strength. I normally take it down a little from the default. Simply put, the closer it is to 0 (left) the more weight it will give to the prompt image, and the closer it is set to 1 (right), the more weight it will give the text prompt. I tend to start just right of the middle so that the AI will ‘understand’ that I want to keep the main element of the image (i.e. Lottii), but will have some freedom to start changing details.

Next, we need to adjust the general settings in terms of steps, guidance scale, negative prompts, etc. This is accessible from ‘settings’.

Though most people assume it’s best to set the steps level to ‘insane’, it’s really not. Honestly, trust me on this, start with it set to ‘medium’ and then work iteratively. Likewise, with the guidance scale, many assume it’s best to set it high. I find it is quite the opposite. Set it too high and you stifle the AI’s ability to ‘have fun’ and be creative. Sure, if you have an extremely specific and detailed prompt, you might want to set it to strict, but for this project, we want to nudge and inspire the AI, so it works with us, rather than fighting it.

So, let's let ‘steps’ to medium and guidance scale to ‘normal.’

We’re not going to worry about seed or negative prompts for this project, though they can be useful. Basically, using the same seed and similar prompts can help you keep consistency with images while adding negative prompts simply helps you avoid certain things.

The scheduler refers to the specific algorithm that removes the noise during image creation. It sounds complicated, but you can normally use the default setting. If you experiment, you’ll also start to figure out which schedulers work best for certain types of images. But, right now, we’ll just leave it on EULER.

Iterative image generation

Righty, we’re good to go, so now the fun starts! I hit the orange button and generated my first image. Wow, that was quick eh! The mage servers must be quiet (well, America is still asleep). And, yes, we’ve already got a nice image. Thanks to the image and text prompts, the AI has already got the general gist of things, and the image looks great in terms of realism. But, it's not exactly what I want - she’s still kneeling in a bath, or maybe a hot tub, but it's not the fun-filled waterpark I’ve got in mind.

So, let’s start nudging things in the right direction. Some people would go straight back to tweaking the text prompts, but I’m not going to do that just yet. I want to see where the AI will go first. So, I’m going to hit ‘Rerun’ a few times and see what it comes up with.

After a few reruns (you can see your image history in ‘Profile’) the AI has come up with some good images, but it’s still not really getting the waterpark thing. So, I’m going to pick the image I like best here and then mask the background.

All I need to do is select the best image, open it, and then select ‘Reimage.’ The original picture of Lottii has now been replaced in the prompt window by the selected new version. Then I just need to open the image and use the inpainting tool. You can automatically select the background, but I find this doesn’t work well, so instead I’ll select the background manually (this is much easier if you’ve got a pen and tablet).

Okay, so that’s the background selected, so we’ve just left the girl and water alone. I’ll hit ‘save’ and ‘ready’ and then crack on with creating a fantastic waterpark.

Now, I’m only going to be editing the selected background area, so I need to change the text prompt. I’m going to go with:

“Large, busy waterpark in summer. Kodachrome-style photograph.”

I could also change the AI model if I wanted to, but I think I’ll stick with CyberRealistic. So, let’s see what happens.

Excellent! The AI has understood and we’re definitely more ‘outside’ and going in the right direction. But, it’s still not the fun waterpark I’ve got in mind. So, let’s give the AI some more freedom. Now I’ll select ‘Remix’ - this will keep the inpainting mask in place but allow us to change the prompt strength. I’ll move the prompt strength to the right a bit, up to 0.7. Remember, the closer it is to 0, the more it pays attention to the prompt image, and the closer it is to 1, the more weight it gives the text prompt. Right now, we are only editing the background, and the most important thing is the text prompt ‘waterpark’.

Let’s run it a few times and see what we come up with…

Okay, we’re getting there, but I think we need waterslides and a splash of colour. So, let’s add some more detail to the text prompt, and change the prompt strength to 0.8.

“Large, busy outdoor Spanish waterpark in summer. Bright yellow and blue water slides. Kodachrome-style photograph.”

Okay, we’re going in the right direction. We’re still not where I want to be, but let’s go ahead and combine the image again and work from this as a base. So, I’ll hit ‘Reimage’.

Now, we’re editing the whole thing again, so I want to change the text prompt. But, let's tweak a few details too. We’ll make the water a nice colour, give her brighter red hair, and change her swimsuit.

We’ll also bring the prompt strength back down a bit so that it is more balanced between the image and text prompts.

“Gorgeous young woman, soaking wet with red hair, wearing a tight swimsuit at a busy outdoor Spanish waterpark in summer. Clear blue water and bright yellow and blue water slides. Intense sunshine. Kodachrome-style photograph.”

Now I’ll run through some more iterations…

We’re definitely getting there! Now I’ve pretty much got what I originally wanted - a redhead woman wearing a swimsuit, having fun at an outdoor waterpark, I can play around with minor details.

Remember, all your iterations are automatically saved, so you can experiment and always go back to a previous one. Try adding detailed prompts and altering the prompt strength balance. Really, you can do this for as long as you want. Sometimes, the smallest detail will make an image, but, be warned, it's possible to get lost in this process and lose many, many hours!

Anyway, when you’ve finally got the image you want, you can select ‘enhance’ to enlarge it. I’d recommend using ‘Upscale only’ rather than the option with ‘Face Fix’ (it tends to lose character and individuality). You can then download the image.

This might seem like a long-winded process, but when you get used to it, it flows very easily, and you can spend as little or much time as you want on it. Of course, you can generate images entirely using text prompts, but as I said earlier, I prefer using a text/image combo - especially because I only use my own original images as a base. It makes the entire process more of a creative collaboration between myself and the machines.

So, here’s the original image of Lottii next to some of my final AI-generated iterations:

As a rule, the more time you spend experimenting, the happier you'll be with the final image, but we all have to cut things off somewhere, so I'll wrap up here. This was just a very basic introduction to one of the techniques you can use to create generative AI images using artificial intelligence tools at mage.space - so, I'll probably be back with more guides soon. In the meantime, enjoy getting creative!