Getting Started with Prompting
The prompt input field is at the top of the screen. Here, you type precisely what you want to see generated.
You can type in just the elements you want to see, or you can type in full sentences – as the system can understand both well. The key though is clarity - the more our AI tools understand your input, the higher adherence, detail and quality you’ll generally get.
Start with the main subject
Begin your prompt by identifying the main subject you want the AI to focus on. For example, if you want the AI to generate art featuring a cat, specify that as your subject. This gives the AI a clear starting point to work with.
Add more descriptions to your subject
Expand on your subject by including additional details such as the location or action. By including descriptors like "colorful" or "sitting in a park" for example: "white cat sleeping inside a box" provides more context for the AI to create a specific visual representation.
How do I create better prompts?
One important note to always remember is that typically words at the start of the prompt are weighted more heavily than those at the end.
Let’s look at some examples:
Say we want to generate a knight on a horse galloping across a stormy beach. But we also want the thunderstorm to play a prominent role and dominate above the knight himself - any of the following three would be acceptable but the first would be potentially most effective, as the storm prompt features first:
”A stormy night sky with lashing rain backdrops a knight frantically riding a horse down a dark beach.”
A knight at night frantically rides a horse down a dark beach, with a stormy sky and lashing rain backdrop”
”Stormy backdrop with rain. Knight riding horse. Dark beach. Night time.”
When comparing the three images, it’s clear that the first is focusing on the storm itself as a priority with bolts of lightning and a longer angle that positions the knight as smaller compared to the vista around them.
In the second image, the knight takes front and centre, with detail being focused there and the storm itself not as chaotic, but rather more alluded to in colour and tone.
The third image has managed to become somewhat more static and painting-like with its slight lack of clarity, focusing on both elements while giving slightly more weight to the backdrop. Whilst certainly not bad, it does lack the dynamic force of either the horse’s movement or the chaotic storm from the other images. This is why a user knowing what they want to create beforehand can allow them to better create their prompts and – by extension – make what they create closer to their ideas.
A good idea with prompting is to start small. The longer sentences get the less likely it is the model will accurately understand, so we advise users to start with a short sentence, or a handful of words, and build outwards.
How do I get more detail?
Simply put users will get more detail out the more they put in. For example, they could use the prompt. “Man sitting on a chair playing violin” and they would get a pretty wide selection of choices due to the freedom the model has been given. Something like this could be expected:
But then if more description is given to create specifics, the model has more to work with, such as: “Old man sitting on a chair playing violin by mountains” which will create this far more specific and prompt adherent image:
This nifty little feature is a great first defence against details that are unwanted (such as nudity, extra limbs or digits). All you need to do is add in what you don’t want to see, and the model is far less likely to include them in the image.
I am still getting elements I don’t want...
If you find you can’t stop unwanted aspects from being produced even with a negative prompt, it’s advisable to change models, aspect ratio, or – with Alchemy – the style, to see if that improves things.
If not, Canvas is a fantastic tool to manually remove unwanted aspects, for which we have a guide here (coming soon).
Can I create style through reference to popular concepts?
Users can indeed create a style by referencing popular works or individuals - such as ‘in the style of Van Gogh, or ‘in the style of starry night painting.’ This can also be movies, games and even studios. For example here is a “Studio Ghibli-style image of a hot-air balloon race,” which clearly emulates the style of the famous animation movie studio. On this aspect – like much with image generation – experimentation is key.
More Tips and Tricks
Utilise modifiers: Consider what form or medium you want the artwork to take. Do you envision a photo, digital painting, 3D render, or a tilt-shift effect? Incorporate these modifiers into your prompt to guide the AI's creative process. For example, you can specify "a photo of a cat sitting on a bench in a park" or "a tilt-shift photograph of a cat sitting on a bench."
Consider the order of elements, especially when using Prompt Magic v2. If you're using Prompt Magic v2, it can be helpful to start your prompt with the most crucial aspect. For instance, if you want a tilt-shift effect, begin your prompt with "Tilt-shift photo of a cat sitting on a bench in a park." This ensures that the AI model understands and prioritizes the key element from the start.
Add more reasonable descriptions: While it's important to be creative and imaginative, it's also crucial to provide realistic and achievable instructions. Avoid prompting something overly complex or unrealistic, as it may result in less visually pleasing or coherent outputs. Try to maintain a balance between creativity and feasibility.
Use commas to separate concepts in longer prompts: As your prompt becomes more detailed and elaborate, it's helpful to use commas to separate different concepts or elements. For instance, instead of writing "A line art drawing of a beautiful cat sleeping on a red bench" try "A line art drawing of a beautiful cat sleeping, on a red bench" This formatting makes it easier for the AI to understand the distinct components of the prompt.
Explore magical words: Research and discover specific words or phrases that can have a transformative effect on the generated art. One famous example is "Artstation". Including such magical words in your prompt can yield unique and unexpected results. Experiment with different terms and see how they influence the artistic output.