The ULTIMATE prompt guide! Make AI easy.

Hey! This is a prompt guide for newbies, explained in a simple way and illustrated with examples. :)

Hello there! I'm Aia. :D

Have you ever tried using an AI art generator and ended up with a cat that looks more like an octopus? Well, that’s probably because of your prompt. Today, we’re going to learn how to write prompts that won’t turn your cat into an octopus… unless that’s what you’re into!

But... what’s a prompt?

A prompt is basically the instructions you give the AI to create an image. It’s like telling a painter, “I want a portrait of my cat”, but in this case, the painter is a bunch of algorithms and code. And it doesn’t care about paint stains! You can check the vidoe version of this tutorial too.

Opening NightCafé

Alright, let's get to it. For this tutorial, I'll be using DreamShaper v8 in NightCafé - it's a free model and I'm running low on credits lol. I'll also be using the default sampler (DDIM, but you don't have to touch anything, as it's chosen by default for this model), and I'll turn off the NightCafé default preset. Click on "Advanced mode" to turn it green, it will disable presets.

You have two options now. You can either click on this link (or the one above Saying DreamShaper v8) and then on "USE MODEL" (if you're new, you'll have to create an account first) or acces NightCafé and click on the create button (usually top right, but if you're new it's in the centre, and you'll have to create an account before continuing). Now, click on the model (SDXL 1.0, just below the advance mmode toggle button). It will open a tab with all the models available. Over the model images, you'll see several buttons: "All", "SDXL", "SD 1.5" and "Other". Click on SD 1.5 and choose the model DreamShaper v8 (the one with the dragon).

Now, just click on "Done" (top right corner; it should close the model selector). You don't have to touch other settings, other than the prompt box, which is just below the chosen model.

Okay. Now you should be ready to create art.

Prompt structure

For writing a good prompt, you should consider the following aspects:

#1 - Subject.

Well, this may be a bit obvious, but deciding what to make is the first step! For this tutorial, we'll be making an elf. Someone new to AI art would write

An elf

And, while it's a good start, the AI can't read our minds (yet!), so providing a few more details would be better. Adding things like some physical details, clothing and what' happening in the image produce more better results. You can also add the place and background. Let's try now with

An elf with silver hair wearing a white tunic casting a spell with a magical staff in the forest

Now, that's much better!

But we can make it even better. Instead of using that many linking words, we're going to separate words with commas.

An elf with silver hair and a white tunic, casting a spell with a magical staff, forest

The differences are small but noticeable. That's because, if we add too many words, the AI will get too much information, and other words will lose importance. The earlier in the prompt, the stronger, so the last words will have a smaller impact on the final result.

If you want to add other secondary subjects, go for it!

#2 - Style and medium.

The next things to take into account are the style and the medium. Do you want your image to be a watercolour painting or a 3D render? Well, now it's the moment to say it! For this example, we are going to make a realistic portrait, as it generates good results, but feel free to experiment with other styles! I'm obsessed with splash art paintings and macrophotography, so don't be scared to try absurd combinations! To separate the different parts of the prompt, we'll be using commas. The prompt would be like this now:

An elf with silver hair and a white tunic, casting a spell with a magical staff, forest, realistic portrait

Well, I think the images are slowly getting better!

You can also add artist names to tell the AI that you want the image to be in their style, as well as art websites, studios and render apps.

An elf with silver hair and a white tunic, casting a spell with a magical staff, forest, realistic portrait by Élisabeth Louise Vigée Le Brun, Unreal Engine 5, Artstation, Artgerm

#3 - Other modifiers.

Modifiers are basically the words used to describe the image. After having made a basic prompt, you can add more modifiers to guide the AI to the result you're trying to get. Some categories you may want to add are:

-Lighting: studio lighting, natural lighting, volumetric lighting... For this example, I'll use natural lighting:

An elf with silver hair and a white tunic, casting a spell with a magical staff, forest, realistic portrait by Élisabeth Louise Vigée Le Brun, Unreal Engine 5, Artstation, Artgerm, natural lighting

-Pose and perspective: How is the main subject in relation to the viewer? What can we see? And how?: centred, head and shoulders portrait, full-body portrait, top view, back view, isometric... I'll use centred head and shoulders portrait for this example:

An elf with silver hair and a white tunic, casting a spell with a magical staff, forest, realistic portrait by Élisabeth Louise Vigée Le Brun, Unreal Engine 5, Artstation, Artgerm, centred head and shoulders portrait

-Adjectives: beautiful, perfect, amazing, awesome, incredible... They can be put with any other words, but I'd recommend using them to emphasise specific parts of your prompt. Colours are also welcome, but I'd recommend you to use as few as possible, as the AI tends to use them for everything (if you say purple eyes and green hat, the most probable result will be both in the same colour).

A beautiful elf with silver hair and a white tunic, casting a powerful spell with a magical staff, dark forest, realistic portrait by Élisabeth Louise Vigée Le Brun, Unreal Engine 5, Artstation, Artgerm, amazing composition

-Photography and detail modifiers: These modifiers are used to make the image more coherent and remove flaws, as well as making it more realistic (sometimes): 8k resolution, sharp focus, intricate details, amazing resolution, bokeh, 64 megapixels

An elf with silver hair and a white tunic, casting a spell with a magical staff, forest, realistic portrait by Élisabeth Louise Vigée Le Brun, Unreal Engine 5, Artstation, Artgerm, sharp focus, intricately detailed, 8k resolution

Alright, now let's put everything together!

A beautiful elf with silver hair and a white tunic, casting a powerful spell with a magical staff, dark forest, realistic portrait by Élisabeth Louise Vigée Le Brun, Unreal Engine 5, Artstation, Artgerm, natural lighting, centred head and shoulders portrait, amazing composition, sharp focus, intricately detailed, 8k resolution

Now, that is another level! Take into account that those images were all generated with the default settings, so the image resolution is not very high. If you want to upscale them, open your creation in NightCafé and hover over the image. You should see a small menu appear at the bottom. Click on the symbol of four arrows pointing outside, and if you don't see it, click on the three points on the right part of the menu, where it says "More", and click on "Enhance". A menu with the upscale sizes will appear. Choose the resolution you want and then click on "Upscale" (bottom right).

And here's the comparison:

I did a 6x upscale, and the result is really noticeable, just look at the eyes, the hair and the staff and compare them in both images.

#4 - Weights.

Some models let us add weights. But... what is that? Well, it's basically a way to give more or less importance to certain parts of the prompt. Note that not all the models have this feature, but those that have it let you know below the prompt box in the advanced mode.

Prompt weighing works by adding parenthesis to the selected words and then the weight. I know this might be a bit confusing, so here's an example:

A realistic anime cat watching the stars over the roof

A (realistic:1.3) (anime:1.1) cat watching the stars over the roof

As you can see, all I did was adding the parenthesis and a number, but the results are quite different. While the first one is realistic with a very little touch of anime details, in the second one it's easier to appreciate the anime details, still keeping it realistic.

That number I added is the weight, letting the AI now how much importance it should give to the text inside the parenthesis. It should be between 0.7 and 1.5, as more might drive the algorithm crazy. 1 is the default weight. So the weighing structure should be (text:weight), where text is the prompt (which can be several words) and weight is the weight number. Don't forget the parenthesis and the colon!

I'll apply this to the elf example:

A (beautiful elf:1.3) with silver hair and a white tunic, (casting a powerful spell:1.2) with a magical staff, (dark forest:1.1), realistic portrait by Élisabeth Louise Vigée Le Brun, (Unreal Engine 5:0.8), Artstation, Artgerm, natural lighting, (centred:1.2) head and shoulders portrait, amazing composition, (sharp focus:1.1), (intricately detailed:1.4), 8k resolution

As you can see, while the result is quite similar to the previous ones, several things have changed. The forest is a bit more defined, as well as the details are somewhat sharper. The elf is in the centre of the picture, and everything looks a bit less 3D-rendered. The image also looks as if she was casting a spell (whispering in the first one and mentally in the second). With the beautiful elf weight I also got the AI to focus more on the elf, getting 5 fingers per hand, as well as decent eyes (which usually are the major flaws of AI images).

So it looks like this is the structure for a good prompt:

Main subject with details

Secondary subjects

Background

Style

Artists

Lighting

Pose and perspective

Adjectives

Details

Note that none of the fields is compulsory, and that I'm not a professional, so there might be other structures that also work. Also, sometimes it's better to write shorter prompts, with less details.

Conclusion

And well, that's everything for today! I hope you found this interesting and helpful! Don't forget to subscribe to my YouTube channel, where I have a video version of this post, and to follow me on NightCafé! Here's the link to a NightCafé collection with the images made during this tutorial.

Until next time, keep dreaming, keep creating, and keep having fun! 🚀

Bye for now!

Aia