I’ve been learning how to translate the images I see in my mind into digital images. I prefer to use the DALLE implementation inside Microsoft Designer because I can use these images in canvases with text and more to create more than “just art”. Note that Bing Image Creator uses the same DALLE engine, thus prompt “engineering” for both is identical. I’ve also used Adobe’s Firefly (Beta), which uses a different engine, but produces striking results. Adobe’s AI image creator offers additional canvas sizes and not just square 1:1. While I can use Runway to expand the canvas for DALLE 1:1 square image, it’s easier to specify the canvas size in advance for composition purposes if I want another canvas size.
Microsoft DALLE Square 1:1 vs. Adobe Firefly 16:9
DALLE 1:1
Adobe Firefly 16:9
With respect to prompt engineering, a simple prompt produces results, but sometimes the image(s) aren’t quite what I had in mind. Using Bing Image Creator, I specified “family of animal plush toys”. I found the output to be satisfactory, but not ‘exciting’ or attention grabbing.
Similarly, Adobe Firefly generated slightly more vivid images, but still, in my opinion, very ordinary and bland. I had to use a different prompt “group of toy plush animals” because the Adobe App found something objectionable and against their standards in the original “family of animal plush toys” prompt I’d used in Bing Image Creator (at least the first few times I tried this language).
So, a little prompt engineering was needed to produce the more vibrant image(s) that I wanted for a specific project. You can see the additional text used in the prompts in the images below.
As for prompt engineering with DALLE, here’s what Bing’s Chat Creator had to say:
DALLE is a deep learning model that can generate images from text prompts. It is a powerful tool for creating art and visualizing ideas, but it also requires some skill and creativity to use it effectively. In this blog post, we will explore what is effective prompt engineering for DALLE AI generated art, how to get the best results, and why a text prompt is improved by prompt engineering.
Prompt engineering is the process of designing and refining a text prompt that guides DALLE to generate the desired image. It involves choosing the right words, phrases, and modifiers that describe the image in a clear and specific way. Prompt engineering also involves experimenting with different variations of the prompt and evaluating the results.
Effective prompt engineering for DALLE AI generated art has several benefits:
– It increases the chances of generating an image that matches the intended vision and style.
– It reduces the chances of generating an image that is irrelevant, distorted, or offensive.
– It enhances the creativity and expressiveness of the image by adding details, emotions, and nuances.
There are no fixed rules or formulas for effective prompt engineering for DALLE AI generated art, but here are some general tips and examples that can help:
– Use descriptive and concrete words that specify the main elements and features of the image. For example, instead of “a dog”, use “a golden retriever with a red collar”.
– Use modifiers and adjectives that indicate the size, shape, color, texture, position, orientation, mood, or emotion of the elements. For example, instead of “a house”, use “a cozy cottage with a thatched roof and a chimney”.
– Use conjunctions and punctuation to separate and organize the elements of the prompt. For example, instead of “a forest with animals”, use “a forest with trees, flowers, and mushrooms; a deer grazing in the clearing; a squirrel climbing on a branch; a bird singing on a nest”.
– Use metaphors, similes, analogies, or references to existing artworks or styles to convey the tone or theme of the image. For example, instead of “a sunset”, use “a sunset like a painting by Monet” or “a sunset that looks like cotton candy”.
– Use feedback loops to iteratively improve the prompt based on the generated images. Try adding or removing words, changing the order or structure of the prompt, or using synonyms or antonyms. Compare the results and see what works best for your goal.
Effective prompt engineering for DALLE AI generated art is both an art and a science. It requires creativity, intuition, and experimentation. It also requires understanding how DALLE works and what it can and cannot do. By applying these tips and examples, you can improve your skills and create amazing images with DALLE.