Image by Stella Sky
Everybody who works with AI has heard of Chat GPT, and so has I. Of course, you can use it for different purposes, but I immediately thought of using it as a text prompt writer for Midjourney.
I registered a new account with my email and phone number on the openai website and started playing around.
ChatGPT: Optimizing Language Models for Dialogue
We've trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for…
openai.com
What is “Chat GPT-4”?
First, I asked my new tool to tell me more about itself.
Me: “How does chat GPT-4 work?”
Chatbot: “GPT-4 (Generative Pre-trained Transformer 4) is a large language processing machine learning model developed by OpenAI. It uses a deep neural network to generate human-like text, and can be fine-tuned for a variety of language tasks such as translation, summarization, and question-answering.
To generate text, GPT-4 takes in a prompt and uses its knowledge of language patterns to generate a response. The model is pre-trained on a massive dataset of text, which allows it to have a wide range of general knowledge and understanding of how language works. When generating text, the model considers the context of the prompt and uses this information to determine what words and phrases are most likely to come next. The generated text is then reviewed by the model to ensure that it is coherent and makes sense.
GPT-4 can be fine-tuned for specific language tasks by training it on a smaller, more specialized dataset. This allows the model to learn more about the specific task it is being trained for and generate more accurate and relevant text.”
Me: “Where is the dataset of text coming from?”
Chatbot: “GPT-4 is trained on a massive dataset of text, which is compiled from a variety of sources. This can include books, articles, websites, and other publicly available texts. The specific sources and size of the dataset are not publicly disclosed, but it is believed to be one of the largest datasets used for language processing to date.
The purpose of using a large dataset is to give the model a wide range of general knowledge and understanding of how language works. This allows the model to generate text that is more coherent and realistic, as it has seen a wide range of language patterns and can use this knowledge to generate text that follows similar patterns.
The dataset is processed and cleaned before it is used to train the model, in order to remove any errors or inconsistencies and ensure that the model is learning from high-quality data.”
If you like my content and want to learn more about AI art and how to make money with it, consider joining Medium as a member through my link. You will be able to read every story on Medium. Thank you so much!
Me: “Are there concerns about copyright and plagiarism?”
Chatbot: “There could potentially be concerns about copyright and plagiarism when using a large dataset of text to train a language processing model like GPT-4. It is important to ensure that any text used in the dataset has been properly licensed and that the appropriate permissions have been obtained.
In some cases, it may be necessary to redact or remove certain texts from the dataset if they are found to be in violation of copyright laws. It is also important to consider how the model will be used, and to ensure that it is not being used to generate text that could be considered plagiarized or in violation of copyright laws.
That being said, OpenAI is a reputable organization, and it is likely that they have taken steps to ensure that the dataset used to train GPT-4 is properly licensed and does not violate any copyright laws.”
Me: “Good to know!”
Chatbot: “I’m glad I could help clarify things for you. If you have any other questions, don’t hesitate to ask!”
How polite! I ask myself how a chatbot can be “glad”, but I guess it has been programmed to give that answer at the end of the conversation.
After this theoretical lesson, I started to test it as a text prompt engineer. This could be a profitable side hustle for you. Read more about this here:
Midjourney’s interpretation of Chat GPT-4 text prompts
- Prompt “Spanish woman holding flowers”
Me: “Tell me a text prompt for Midjourney to create a portrait of a young Spanish woman, holding red roses, with perfect hands!”
(If you have worked with any AI art generator, you know why I ask for perfect hands…)
Chatbot: “The young Spanish woman stood proudly, her long black hair cascading down her shoulders. She held a bouquet of red roses in her delicate hands, the perfect complement to her olive skin. As she gazed confidently at the canvas in front of her, the artist knew that he had to capture the essence of her beauty in his portrait. With careful brushstrokes, he began to paint her striking features — the almond shape of her eyes, the graceful arch of her eyebrows, and the fullness of her lips. But it was her hands that he found most captivating, with their slender fingers and perfectly manicured nails. As he worked, he knew that this would be a portrait to remember, a masterpiece that would showcase the young woman’s beauty and spirit for all to see.”
Midjourney’s image with
“ — q 2 — ar 2:3 — style 4b — stylize 800 — upbeta — v 4”
Image by Stella Sky
Beautiful, but no hands on all four images. Probably MJ didn’t want to mess up my first GPT-4 image.
Let’s try something different:
2. prompt “Futuristic spaceship”
Me: “Tell me a text prompt for Midjourney to create a futuristic spaceship with many details floating in space. It should look photo realistic.”
Chatbot: “The spaceship hovered in the vast expanse of the galaxy, its sleek silver exterior reflecting the bright glow of the stars. From a distance, it seemed almost ethereal, a mirage floating in the darkness. But as the viewer moved closer, the ship’s many intricate details came into focus. The smooth curves of its hull gave way to rows of portholes, each one a window into the bustling activity within. Antennae and sensors bristled from the ship’s surface, sending and receiving signals from far-flung corners of the cosmos. And at its heart, the ship’s engines pulsed with energy, ready to propel it to even greater heights. This was a vessel unlike any other, a masterpiece of futuristic design, and as it drifted through the void of space, it seemed almost alive, a living, breathing entity exploring the unknown.”
Midjourney’s image with:
“ — q 2 — ar 3:2 — style 4b — stylize 900 — chaos 70 — upbeta — v 4”
Image by Stella Sky
3. prompt “Something futuristic” (I like to be surprised!)
Me: “Tell me a futuristic text prompt for Midjourney.”
Chatbot: “The city of the future was a bustling metropolis, a vast and vibrant hub of activity. Towering skyscrapers reached towards the sky, their sleek glass and steel facades gleaming in the artificial light. High above the streets, sleek flying cars zoomed through the air, navigating their way through the crowded airspace with ease. And on the ground, sleek robotic assistants and delivery drones moved about, performing a never-ending array of tasks and errands. The city was a place of endless possibility, a place where technology and humanity coexisted in perfect harmony. And as the sun set on the horizon, casting a golden glow over the cityscape, it was clear that this was a place where anything was possible, a place where the future was now.”
Midjourney’s images with:
“ — q 2 — ar 3:2 — style 4b — stylize 900 — chaos 70 — upbeta — v 4”
Image by Stella Sky
Really interesting how the two AIs work together. I usually don’t write such long descriptive text prompts, even if I use the “Prompt builder”.
More ideas and text prompts!
I continued working with both tools and was rewarded with stunning images full of intricate details and unexpected results. Over 200 images were created with Valentine’s postcards, steampunk backgrounds, angels & devils, cyberpunk models, and incredible anthropomorphic animals.
Image by Stella Sky
Soon I will publish stories for different ideas and prompts made with Chat GPT-4 and Midjourney and give you more tips. Did you already use both tools together?
Much love,
Stella Sky
