How to Use ChatGPT to Generate Prompts for Stable Diffusion -

How to Use ChatGPT to Generate Prompts for Stable Diffusion

Creating text-to-image models such as Stable Diffusion needs to have good and high-quality prompts. If you don’t know how to craft a good Stable Diffusion prompt, ChatGPT can help you but it needs some special methods to get the best results.

Here is the step-by-step guide on using ChatGPT to generate Stable Diffusion Prompts.

What is ChatGPT?

ChatGPT, which was launched by OpenAI in November 2022, is a chatbot that utilizes the GPT-3 series of Large Language Models. These models were trained using text data from the internet containing hundreds of billions of words. To put this into perspective, a book with hundreds of billions of words would take over 5,000 years for a person to read out loud. As a result, ChatGPT possesses a vast amount of human knowledge and is capable of generating essays, poems, works of fiction, and even computer code.

You may create an account with AI-PRO.org to access AI tools powered with ChatGPT here.

In What Manner Was ChatGPT Trained?

ChatGPT was trained on top of GPT-3.5, a powerful large language model that functions as an autocomplete tool. By predicting the most probable next words based on a few words or sentences entered, GPT-3’s autocomplete feature is enhanced by the vast amount of data it was trained on. This enables it to generate coherent and mostly accurate content, sometimes even an essay.

During the training phase, human trainers played a vital role in fine-tuning ChatGPT’s responses. The model was programmed to provide answers that mimic those of human trainers. Additionally, the trainers ranked ChatGPT’s answers to prioritize responses that humans prefer.

Does ChatGPT Know Stable Diffusion?

ChatGPT does not know anything about Stable Diffusion. When asked “What is Stable Diffusion,” ChatGPT provides a different answer:

“Stable diffusion refers to a type of diffusion that is characterized by a state of equilibrium or balance between the concentration of particles or molecules in different areas. In other words, the rate of diffusion is constant and there are no significant changes in the concentration gradient over time.”

However, ChatGPT knows text-to-image models such as Dall-E. You need to take note of how specific your prompt is in order for ChatGPT to create the correct response, particularly in asking for a Stable Diffusion Prompt.

How to Use ChatGPT?

To initiate a conversation with ChatGPT, visit your ChatGPT account and type your question or prompt in the input box located at the bottom.

Asking for Illustration Prompts

This section will be testing ChatGPT to provide prompts for illustration portraits, which is one of the hardest to get accurate results.

Close-Up Illustration

To generate high-quality prompts for Stable Diffusion using ChatGPT, there are three important techniques to follow:

  • Specify the image style by requesting a particular type of image, such as a “photorealistic close-up illustration”.
  • Fix the subject of the prompt, in this example, the actress “Selena Gomez” to ensure consistent keyword use and influence on style and composition.
  • Prime the prompt generation process by providing two example prompts.

Putting these techniques together, the ChatGPT prompt would go something like this:

“Provide a high quality prompt for stunning close-up photorealistic illustration of Selena Gomez for text-to-image models (Stable Diffusion, midjourney or Dalle2), such as this one:

  • Keanu Reeves portrait photo of an Asia old warrior chief, tribal panther make up, blue on red, side profile, looking away, serious eyes, 50mm portrait photography, hard rim lighting photography–beta –ar 2:3 –beta –upbeta –beta –upbeta –beta –upbeta
  • A vibrant professional studio portrait photography of a young, pale, goth, attractive, friendly, casual, delightful, intricate, gorgeous, female, piercing green eyes, wears a gold ankh necklace, femme fatale, nouveau, curated collection, annie leibovitz, nikon, award winning, breathtaking, groundbreaking, superb, outstanding, lensculture portrait awards, photoshopped, dramatic lighting”

Here are ChatGPT’s Responses, and how they appear on Stable Diffusion:

ChatGPT Prompt #1:

A close-up photorealistic illustration of Selena Gomez with soft, natural lighting, showcasing her natural beauty and radiance. Selena should be shown in a relaxed and intimate pose, with a subtle smile and gentle gaze directed towards the viewer. The focus should be on her eyes, which are highlighted with a soft, warm glow, and her hair, which cascades over her shoulders and frames her face. The illustration should be in full color, with intricate attention to detail and texture to bring out the depth and richness of Selena’s features.

ChatGPT Prompt #2:

A high-contrast photorealistic illustration of Selena Gomez with dramatic, stylized lighting. Selena should be shown in a bold and powerful pose, with her head turned to the side and her eyes gazing off into the distance. The focus should be on her strong facial features, with sharp shadows and highlights emphasizing her bone structure and contours. The background should be dark and moody, with abstract shapes and patterns that complement Selena’s dynamic presence. The illustration should be in black and white, with crisp lines and graphic elements that create a striking visual impact.

ChatGPT Prompt #3:

A dreamy, surreal photorealistic illustration of Selena Gomez in a fantastical, otherworldly setting. Selena should be shown in a whimsical and ethereal pose, with a serene expression and a dreamy gaze directed upwards. The focus should be on her delicate features, which are accentuated by soft, diffused lighting and a hazy, dreamlike atmosphere. The background should be an abstract, fantastical landscape that transports the viewer into a magical and enchanting realm. The illustration should be in full color, with a soft, painterly style that captures the enchanting mood of the scene.

Full-Body Illustration

If you want to ask ChatGPT for a full-body illustration, you can ask for a follow-up question and add “full-body” to your question. Here are some example prompts to ask ChatGPT:

“Give me 2 sample prompts for stunning photorealistic full body illustration of Selena Gomez for text-to-image models (Stable Diffusion, midjourney or Dalle2). Use these examples as format:

  • full body portrait of beautiful happy Selena Gomez, ethereal, realistic anime, trending on pixiv, detailed, clean lines, sharp lines, crisp lines, award winning illustration, masterpiece, 4k, eugene de blaas and ross tran, vibrant color scheme, intricately detailed
  • clear portrait of a superhero concept between spiderman and batman, cottagecore!!, background hyper detailed, character concept, full body, dynamic pose, intricate, highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, art by artgerm and greg rutkowski and alphonse mucha”

ChatGPT Prompt #4:

A photorealistic full-body illustration of Selena Gomez as a warrior princess, with intricate armor and weapons, in a dynamic pose, set against a detailed fantasy landscape. The illustration should be highly detailed, with crisp lines and vibrant colors, and should showcase Selena as a powerful and confident warrior.

ChatGPT Prompt #5:

An ethereal and dreamlike full-body portrait of Selena Gomez as a mermaid, floating in a sea of vibrant colors and surrounded by intricately detailed sea creatures. The illustration should be highly detailed, with clean lines and sharp focus, and should capture the beauty and magic of the underwater world. This illustration should be a true masterpiece, with a stunning color scheme and a level of detail that will leave viewers speechless.

Ask For Other Suggestions

Enhance and adjust the prompts ChatGPT created based on the type of results you want to get from Stable Diffusion. If you liked the prompt that ChatGPT gave, you can also ask for more prompts of the same style by typing a follow-up question like “Can you give me more examples of the first prompt you gave?”

Takeaway

Using ChatGPT to generate Stable Diffusion prompts can lead to better and more accurate results in text-to-image models. As text-to-image models continue to advance, the demand for high-quality prompts will continue to rise. By using ChatGPT, we can generate prompts that will help drive the advancement of these models.

Remember to specify the image style, fix the subject of the prompt, and prime the prompt generation process by providing at least one or two example prompts. This will ensure that ChatGPT produces prompts that are consistent and relevant to your project.

By following these steps and incorporating ChatGPT into your digital art process, you can create effective text-to-image models that meet your specific needs. So why not give it a try today and see what kind of prompts you can generate with ChatGPT?