Prompting is a critical part of getting great results, but it can be quickly learned with a little guidance. If that's what you're looking for, you're in the right place. In this article, we will explore the key elements required for creating great images:

Let’s do it!

Prompting best practices such as adding the most important words at the start of the prompt also applies here. If you haven't read it yet, check out our Prompt Quickstart Guide which explains these prompting essentials.

One of the most important decisions when prompting is choosing a model that will best serve your purposes. Starting out with the right model will save a lot of time from trying to find the perfect prompt. A model that is aligned with your intended style or goals will provide better quality results with fewer keywords, and allow you to focus more on fruitful exploration, rather than small iterations.

For instance, a photorealistic model will generally create images that look like photographs, without the addition of keyword terms like: photorealism, photography, camera type, etc. Though these may still help and influence the prompt, they would be less required than in the case of a general purpose model.

PicFinder currently uses open source image generation models. These are Stable Diffusion models, which are fine tuned by the AI community to deliver better results in an area. Broadly, we have three model categories:

Photorealistic models, like Absolute Reality are trained to create images that look like real photographs. In contrast, fantasy models, like RevAnimated, are more fine tuned on illustration and concept art, and tend to create high quality images with a distinct animated feel. General purpose models are more flexible and can serve a variety of styles, but need to be prompted very specifically to deliver a style, by using adequate keywords in every prompt.

Here’s a comparison of how two different models render the same prompt: close-up portrait of man hiking on forest trail, carrying backpack, mountain scenery in the distance, high resolution photography, sharp focus, perfect features, symmetric face

Photorealistic model - Absolute Reality

Animated model: RevAnimated

Coming soon: SDXL

Thankfully, you can always experiment and generate images with different models. So even if your needs evolve and you find the style you like is different than the one you started with, it’s always possible to easily switch with PicFinder.

Once you’ve experimented with models, the key next step is structuring your prompt. A typical prompt structure should be:

Here’s an example prompt following this structure: a woman wearing an overcoat (subject), walking on street (background), portrait photography (type), full body (style), stock photography (artistic reference), high resolution, sharp focus, 4k, perfect face, beautiful features (resolution and quality), evening light (lighting or color scheme). Preview of the results is below and you can also see it in action here.

Not necessarily. For instance, the subject may not always be a character you can clearly describe, or the importance of elements might be different. You might not care at all about the background, for instance if you are aiming to create print-on-demand art and want to remove the background altogether. But you might care a lot more about the visual style, and add more words to capture it. Here’s such an example.

Prompt: great wave (subject), illustration (type), anime, manga, minimalism (styles), sticker, icon (artistic references), round, white background (background). In this case, the background is not as important, so it’s placed near the end of the prompt. Also, the lighting details are missing, also because they are not a key factor in this image. However, the visual style and artistic references are important, so more words are used to describe them. See more results for this prompt here.

When using keywords to prompt, remember that AI models are trained using image tags found on the internet. Therefore, what we refer to as a beautiful photo when speaking may be tagged on the internet with terms like fashion photography, portrait photography, studio lighting, the name of the photographer, the brand of the camera, etc.

No matter what style of image you are trying to create, considering what technical terms are used to describe that style online can be very helpful. Performing a quick search for prompt examples in that style can also help. But another very handy tool is: the magic prompt.

The magic prompt is an AI tool that suggests words to add to your prompts, which can get you closer to your intended result, or surprise you with entirely new capabilities. My favorite way to use the magic prompt is to explore what is possible within a topic or style, and to discover keywords I might otherwise not think of.

You can access the magic prompt by clicking on the little wand icon to the left of the prompt bar, after you’ve performed a search. For the magic prompt to work, some text needs to already be added to the bar.

The magic prompt will suggest different keywords every time you try it, just like the image generation creates different results. Here are some examples of results I got by using the character prompt shared above, which depicted a woman on a city street.

These results are quite different from my initial prompt but they are also stunning. Now I have new prompt keywords which the magic prompt added. I can add or remove them to different generations to see their visual impact, then use ones I like most in my future prompting. It’s a great way to discover new keywords and expand your prompting knowledge.

You can upload an image you already have or generated, for instance if you are trying to generate characters, you could use the image of a character you already have, or an image that represents the visual style you are trying to achieve. To upload, simply click on the image icon to the right of your prompt bar.

When an image is uploaded, the generation algorithm tries to follow the same structure, but generates entirely new content. The algorithm may interpret the image differently than we do. For instance, it might recreate the layout or the colour scheme, not necessarily a specific item that exists in the image. We are however introducing new features soon to allow you to have more control over which elements you keep from a particular seed image.

Here’s an example where I used a seed image and a prompt to create a children’s book illustration, in a style close to that of Beatrix Potter.

First, here is the result using prompt only, but mentioning the style in the prompt itself:

And here is the result when I also used the same prompt and a seed image. You can see that the color scheme is a lot lighter, the images are brighter, and more of the edges are now blurred, and the illustrations are generally closer to the style than they were before.

The seed image:

The result, which you can also see here:

If you have a series of images you already like, you don't have to start your prompt writing from scratch. You can upload an image and opt for image-to-prompt. An algorithm will analyze the image, and return a series of keywords used to generate similar images.

This feature is extremely helpful when you're working with a new style and you're not quite sure where to start.

These two features are incredibly powerful when combined. They allow you to both get the prompt behind an image, but also set it as an example. Then, you can recreate similar images which will match the subject and the style, but will also contain a good level of variety. You can also introduce changes by prompting, using the best practices from above.

Here's the image we started with above, for image-to-prompt. It's a stunning, AI generated image.

Here are the results obtained by simply uploading the image, running image-to-prompt, and then clicking on the prompt to generate results - which also sets the image as a seed. The results include the same character, colour scheme, and details, for instance the rainbow mane, but they are varied enough to generate fully original results.

We're constantly working on bringing in more capabilities into PicFinder, to create even better images, make prompting simpler, and also allow you to explore and generate images more easily. In the meantime, please vote for features to let us know what truly matters to you, so we can direct our efforts accordingly.