Tag: Stable Diffusion

What is GPT-3?

GPT-3 (Generative Pre-trained Transformer 3) is a large-scale language model developed by OpenAI. It is a deep learning-based natural language processing (NLP) model that uses unsupervised learning to generate human-like text. GPT-3 is trained on a massive amount of text data from the internet and is designed to predict the next word in a sentence. It can also generate short passages of text, which can be used for summarization and question-answering. GPT-3 is the largest and most powerful language model ever created and is capable of understanding, generating, and manipulating natural language.

GPT-3 can be prompted to generate many realistic answers to chat-based input, including responding in poem, with factual information, and with usable code. There is still a long way to go before the model is perfect, as it’s poems aren’t great, will often confidently state incorrect facts and will provide code that doesn’t run. Despite these shortcomings, GPT-3 is a huge stepping stone in the direction of incredible AI tools to improve society.

Learn more by visiting their site: OpenAI

How can GPT-3 create incredible AI Image Prompts

GPT-3 can be prompted to output a set of words that describe the image similarly to how we would prompt for an image. Although the AI doesn’t have an idea of which words exist in the model, due to the vast dataset of Stable Diffusion, often we’re able to create great looking art.

In this example, we’ll be using MitchJourn-E to create the Stable Diffusion images and the beta app for GPT-3 Chat.

To get GPT-3 to create these prompts, I’ll start by specifying “Using only nouns and adjectives”. That should eliminate any connecting words that aren’t as important in image prompting. Then I’ll ask “describe an incredible work of art”, which will start returning the output to plug into our image generation apps. Sometimes it will only output a few words, so I’ll also put at the end “Using 50 words or as many words as possible”, something to get it to output some more useful words.

Examples:

GPT-3 prompt:

Using only nouns and adjectives, describe an incredible and detailed painting of a landscape with as many words as possible.

Output 1:

Majestic mountains tower over a sprawling, verdant valley. A crystal-clear river winds its way through the landscape, cascading over rocks and boulders as it flows towards a distant, azure sea. Wildflowers dot the rolling hills, and a single, graceful tree stands tall on a grassy knoll, its leaves rustling softly in the gentle breeze. The sky is a deep, rich blue, with fluffy white clouds drifting lazily overhead. The whole scene is bathed in the warm, golden light of the setting sun, casting long, surreal shadows across the landscape. It is a truly breathtaking and awe-inspiring sight.

GPT-3 prompt:

Using only nouns and adjectives, describe an incredible and detailed painting of a landscape with 50 words.

Output 2:

Mountains, cliffs, sky, clouds, sun, trees, grass, river, rocks, waterfalls, wildlife, flowers, colors, beauty, serenity, majesty, vastness, depth, perspective, realism, skill, talent, artistry, wonder, amazement, awe, admiration.

Output 3:

Vibrant, colorful sky stretches above rolling hills and tranquil lake. Majestic mountains tower in the distance. Lush, verdant forests and fields dotted with wildflowers. Sun shining, birds singing. Tranquil scene captured with stunning detail and precision.

GPT-3 prompt:

Using only nouns and adjectives, describe an incredible and detailed abstract painting inspired by a landscape with 50 words

Output 4:

Vibrant hues, sweeping brushstrokes, and majestic peaks form an awe-inspiring abstract landscape painting. Lush forests, sparkling waters, and towering mountains are captured in a mesmerizing display of color and texture. A sense of majesty and wonder permeates the artwork, drawing the viewer into a world of beauty and inspiration.

Additional Prompt Info

Negative prompt:

lowres text error cropped worst quality low quality normal quality jpeg artifacts signature watermark username blurry artist name deformed disfigured poorly drawn out of focus censorship amateur drawing bad art poor art messy drawing portrait person people text signature watermark frame framed display

Steps: 50

One of the seeds: 902193268

Width: 640

Height: 384

Prompt weight (CFG scale): 7.5

Sampler: k_euler_a

Output Images

We can use these GPT-3 outputs and plug them into MitchJourn-E to create some beautiful images:

Output 1:

Output 2:

Output 3:

Output 4:

Create your own beautiful art

MitchJourn-E is a free and open source tool available to download on GitHub. Paste your GPT-3 prompts directly into the interface and start generated the most AI AI images you’ve ever seen!

Download now from GitHub: https://github.com/MitchOkabe/MitchJourn-e

Simply put, prompt engineering is the process of carefully crafting and tuning prompts for AI generation. A prompt is a piece of input that an AI system uses to generate output, such as a text, image, or audio clip. By carefully designing and fine-tuning these prompts, we can help AI systems to generate more accurate and engaging outputs.

Now, you might be wondering why prompt engineering is so important. After all, don’t AI systems generate whatever you type into them? While it’s true that AI systems can generate stunning outputs by simply typing whatever you wish into an input prompt, the results can often be unstructured and unpredictable. By providing carefully designed prompts, we can guide the AI system and help it to generate more coherent and useful outputs.

For example, imagine that you’re building an AI system that’s capable of generating pictures. Without prompts, how would it generate a scene of “A statue on the side of a cliff in Japan”? It may choose to generate a beautiful statue, then put the cliff in the background to the side, and the setting is Japan. Or it may have been trained that pictures of Japan, that are by cliff sides, are usually accompanied by Japanese architecture, so a traditional Japanese building becomes a focal point instead of the statue. However, by providing carefully designed prompts, such that different concepts are more or less weighted based on how the model is trained, the AI can create incredible and well imagined art.

Prompt engineering is an essential aspect of AI generation. By carefully designing and fine-tuning prompts, we can help AI systems generate more accurate and engaging outputs. Without prompt engineering, AI systems will generate results based on how the data was trained and will create biases in the resulting work. Prompt engineering allows us to guide the AI system and create more coherent, useful and artful outputs.

Start Creating, like a prompt engineer!

MitchJourn-E is a tool that incorporates Prompt Engineering as a core concept when creating images. You can define weights for each part of your prompt and add helper words and negative prompts automatically. MitchJourn-E is free and open source on GitHub: MitchOkabe/MitchJourn-e: Windows desktop application for Stable Diffusion (github.com)