Contents
- What is a prompt?
- The importance of correct prompts
- How to create prompts
- Using prompts for images
- Using prompts for texts
What is a prompt?
A prompt (from the English "prompt") is a request sent to a neural network in order to obtain the desired text or visual result. The more precisely and clearly the request is formulated, the higher the likelihood of receiving a relevant response. In recent years, marketers have actively integrated neural networks into their work, using them to create graphics for social media, advertising materials, and content plans. To make the most of the capabilities of neural networks, it is essential to master the art of crafting prompts.
The importance of correct prompts
Neural network-based services are becoming increasingly popular, making it easier for users to perform routine tasks and allowing them to save time and resources. For example, in the email builder Unisender, you can not only create images but also translate text or come up with a subject line and preheader thanks to built-in AI-based tools.
There are two variants for writing the term: "prompt" and "promt." From a translation perspective, the correct usage is "prompt," however, due to the complexity of writing and pronouncing it, many users prefer the simplified version.
How to create prompts
For working with neural networks, it is best to use the English language, as most of them were trained specifically in this language. If you send a request in Russian, the service may misinterpret it, which will often lead to an unsuccessful result. For example, the request "A girl looking at the sunset" in Russian might yield only images of girls, while a similar request in English will give a precise result.
If you are unsure of your level of English, we recommend using an online translator like Deepl, which is considered more accurate compared to other alternatives. Prompts can be created in two ways:
- Creating your own requests.
- Using ready-made templates.
The second method can significantly save time and quickly achieve the desired result. For example, by replacing "Russian" with "Chinese" in a template request, you can change the resulting image without having to completely rework the request.
Using prompts for images
Among the popular neural networks that generate images from text descriptions are:
- Midjourney
- Stable Diffusion
- DALL-E
- Craiyon
- Starryai
- GauGAN
- Lexica
- Dream
- Kandinsky 2.0
- "Masterpiece"
The algorithm for working with these services is generally similar and includes several steps. First, you need to specify the object that will be the central element of the image, and then add additional characteristics such as color, properties, and actions. For example, the request "car, golden, racing through the city" will lead to the creation of corresponding graphics.
It is important to remember that requests should not consist of just one or two words—such brevity can lead to the neural network misinterpreting the request. It is better to combine several objects to achieve a more interesting result. For example, creating an image where a raccoon is in a cyberpunk style may yield an unexpected and original result.
Using prompts for texts
The following platforms are most commonly used for text generation:
- ChatGPT
- Bing AI
- YandexGPT
- TurboText
- Retext.ai
- Gerwin AI
- ruGPT-3
Requests for text neural networks differ from prompts for images. They resemble an expanded monologue or a technical assignment. When crafting a request for ChatGPT, for example, it is important to define the role of the neural network, clearly state the goal of the text, describe the context and target audience, and indicate the desired style and format of the response.
To get the most relevant text, it is necessary to formulate the request in detail and clearly, avoiding professional jargon and ambiguous phrasing. This will help prevent receiving vague and uninformative texts.