Artificial Intelligence 7 – Creating images made easy

Now that you know how to write effective prompts, let's take a break from text for a while and apply what we've learned to creating images. Use AI to create images that don't yet exist and that would be impossible or extremely difficult to produce by other means.

Let's start with a simple introduction here as well. For this, you will need the paid version ChatGPT-4. ChatGPT usually creates images in the webp file format. This is suitable for use in online applications. However, many browsers cannot display images in this format. You will need to convert them to jpg, png or a similar format. There are freeware programmes or inexpensive software solutions available for this purpose. This depends on the hardware and software you are using. If in doubt, ask ChatGPT which image conversion software it currently recommends.

We told ChatGPT-4: Create a picture of a monkey. The image above was then generated by AI. As you can see, in the absence of further specifications, an image is generated that is square, with the monkey sitting in the jungle. You did not specify any requirements regarding the breed or appearance, pose, style, etc. The image was created by ChatGPT deciding what would be depicted in detail.

The only additional requirement for this image was: The monkey should eat a banana. We were lucky that the same monkey was depicted here. This is not a given, as the AI itself does not know which image elements were processed in the image and cannot necessarily access them in a further prompt. But consider the effort that would have been required to create this image without ChatGPT. Without an expensive airbrush artist, it would have been virtually impossible. Cost: easily several thousand euros.

Another example. The requirement was: Create an image of a mouse biting into a piece of cheese.

Here, too, no further specifications were defined, so ChatGPT had free rein. But what happened? The manga mouse has two tails!!! Image errors are always possible, and you must always check created images carefully to ensure that all image elements are correct and logical.

The error has been corrected in this image. The task: Create an image of the mouse, but with only one tail. The mouse now actually has only one tail. But it is also a different mouse, as can be easily recognised by its much smaller eyes and significantly smaller nose.

After these brief examples, we need to make a fundamental observation about image generation. ChatGPT can access a lot of digital content. You would say that a human being has been around a lot. Of course, this is not the case for a chatbot. But it can access a significant portion of human knowledge. So, the bot could easily present you with the Mona Lisa in response to the prompt, „Create an image of a smiling, dark-haired young woman in the Renaissance style. The background should suggest a landscape with winding roads and bridges. The image should be almost photorealistic.“ You really need to internalise this! When generating images, you always have the smouldering fuse lying on the table near the black powder. Or you have applied the screwdriver to open Pandora's box. Image rights are the property of the author or creator of a work. This can be a painter, lithographer, graphic artist or photographer. Image rights are always a sensitive issue when it comes to images. Violating them can be expensive. This should not spoil your enjoyment of image generation. But you should be aware of what you may be getting yourself into if you do not take certain precautions.

Now the question for the chatbot: What are the basic considerations when using prompts to create images?

ChatGPT: When creating prompts for image generation, there are a few important aspects to consider in order to achieve accurate and convincing results:

Clarity and precisionBe as specific as possible when describing what you want to see. Vague or unclear prompts will lead to inaccurate results.
Relevant detailsMention specific elements such as surroundings, objects, characters, clothing, colours and moods to create the desired image.
Structure and sequenceStart with the most important element and then add the details. This helps the model recognise the focal point of the image.
Avoid contradictionsEnsure that the description is consistent and does not contain any contradictory instructions.
Style and atmosphereIf you are aiming for a particular style or atmosphere, provide references to well-known styles, periods or cultural influences.
Variations and flexibilitySince the models are not perfect, it may be helpful to try out different variations of the prompt or to allow for a certain degree of flexibility.
Ethics and respect: Be careful not to create discriminatory or disrespectful content. Observe copyright and intellectual property rights as well as personal data protection rights.

With these tips, you can create effective prompts that lead to compelling and accurate images.

______________

Now let's put this into practice: a new task, this time with much more detailed requirements. Here is the prompt:

Create a landscape-format image with a resolution of 1792×1024 pixels. The image should be in colour, with a wide colour spectrum and high resolution. The scene takes place in a modern, brightly lit office with large windows. In the foreground is a long desk with a laptop, several monitors and a few books. The laptop has a black casing, and the monitors display complex data analyses. In the background, there is a large whiteboard with diagrams, notes and various drawings. Next to the desk is an ergonomic office chair. A potted plant sits near the window, and sunlight shines gently through the room. The atmosphere is bright and motivating, with a touch of professionalism.

This prompt describes details, proportions, desired format, colourfulness and colour space in detail in order to generate a clear image using AI. No sooner said than done. ChatGPT delivers. There is still plenty of room for creativity. But the client's requirements are all immortalised in the image. What is not possible: you cannot influence individual image details retrospectively. So you can forget about instructions such as „Remove the whiteboard in the background“ or „Please put a different office chair on the right“. ChatGPT cannot do that. Or perhaps we should say not yet. However, if you still want to influence the created image in this way afterwards, you can certainly ask ChatGPT to do so, but in this case a completely different, new image will be generated.

Now you can entrust your image requests to ChatGPT. However, to minimise any trouble regarding possible copyright infringement, always perform a reverse search on Google.. To do this, open a new Google browser, click on Images and drag your ChatGPT-generated image into the search field. You may need to convert the image generated by ChatGPT into another file format, such as jpg. There are various freeware programmes available for this, depending on your operating system and browser. After the usual cookie queries, you will then receive suggestions of identical or similar images that Google has found, allowing you to assess whether your image already exists. The rule of thumb is: the more detailed the image you request via prompt, the less likely it is that ChatGPT will present you with an image that it has already found somewhere else.

Now it's time to start the adventure of image creation. Remember: use prompts that are as detailed as possible and do a reverse image search on Google. Good luck! By the way: all the cover images for the seven previous posts on artificial intelligence 1 to 7 are, of course, AI-generated. What else?