Artificial Intelligence, AI Art & Prompt Engineering
Improve your AI text-to-image prompts with enhanced NLP
AI-Generated Art: How to Narrativize a Prompt for Stunning Art
Text-to-image prompts are often seen as very clinical and detached. It’s one reason why critics refuse to see it as art (an opinion I disagree with) — as if AI art is just ordered Subway-style from a string of random descriptors:
hyper-realistic, cinematic, detailed, hypermaximalist, octane render, unreal engine 5, 8K
And yes: you can construct a prompt that way, and in another post, I’ll show you how to reverse-engineer a prompt from these generic terms.
But there’s more to it than that. Harnessing the power of natural language processing (NLP) is key to creating far more compelling imaginative art.
With Machine Learning, NLP is our paintbrush
Recently I’ve written about how word choice can give us vibrant colors:
AI is actually very good with mood and tenor; something we forget with prompts. AI is sensitive to subtleties of language. “Shady” is different to “shadowy” is different to “sepulchral”, and this is reflected in art output
Jasper Art is basically Dall-E 2 but the results are different — and better, I’d argue — because Jasper Art prioritizes NLP (natural language processing). Accordingly, it can handle complex sentences with semantic articulation.
Remember, Jasper AI was primarily a copywriting tool before it branched out into the art generator game. This means prompts will respond to tone of voice. The language choices we make — not just the individual words, but syntax and structure — become another moving part or piece of the puzzle.
Using NLP to prompt better AI image generation
OpenClip is about to completely upend how many people prompt. Say goodbye to “Greg Rutkowski”. Dropping names as prompt terms will no longer work in Stable Diffusion 2 (for copyright and ethical reasons). We will have to innovate in how we describe the art we are referring to — and dare I say it — we’re going to have to be more artistic and knowledgeable of art.
One of the directions I’ve been pioneering is “narrative prompts”. This is where instead of a combination of terms strung together at random, we think about what we’re writing and use a more expressive description.
Effectively, paint with words, like authors and poets do, and try to extract a “mental picture” from the AI, just as one would from human imagination.
How do I narrativize a prompt for AI-art generators?
Being expressive with our language won’t come naturally to every prompt engineer. Some of us aren’t creative writers or have language difficulties.
Obviously, this was something I’d have to innovate around, and AI again presented the solution. A while back, I wrote a Jasper recipe that would create images from long text (like a blog post) to illustrate that content:
Building on this, what if I came up with a command that could turn a basic key term-laden prompt into a thing of narrative beauty? A more creative, narrative-driven approach that informs the image? Here’s how it works:
Using AI text to improve your AI image prompt
This command works in Jasper AI Boss Mode, and instructs the AI:
> rewrite the above prompt in entirely new evocative words
Let’s give it a test run.
Here’s a perfectly good prompt adapted from my friend and colleague, Kris Kashtanova (who attained the historic first copyright for an AI comic book).
tiny lion cub, 8k, kawaii, adorable eyes, pixar style, winter snowflakes, wind, dramatic lighting, pose, full body, adventure, fantasy, renderman, concept art, octane render, artgerm
I ran it through my “Narrativize Your Prompt” Jasper recipe and got this:
Lion cub, small but mighty, with eyes that seem to pierce your soul. In a winter wonderland, he stands tall against the snow, wind ruffling his fur. He seems almost like a creature of legend, ready for an adventure. The lighting is dramatic and striking, and the render is breathtakingly beautiful.
It’s much more fleshed out and fully realized. Let’s see what the art generator does with this input:
For comparison, let’s prompt the original version to see the difference:
Addendum: a few people have asked about the omission of “Pixar style” in the narrative prompt, and questioned whether this accounts for the remarkable difference. Good question, and thank you for the opportunity to answer it:
You can see the difference. Many of the reference points are identical in the two prompts, but by giving a more lyrical, evocative description, with full sentences, we got a much more detailed and fully realized image. You’ll notice we didn’t even have to use stale terms like “photorealistic” or “highly detailed” (my pet peeve in prompt engineering) to get these lifelike results.
I think it’s worth the little extra effort.
I hope this inspires you to write longer, more evocative prompts when using AI image generators. With the help of natural language processing, you can generate sophisticated images without having to resort to default keyword combinations can sometimes be lacking nuance and personality.
So next time you’re prompting AI art, take a minute and consider how your words could be interpreted. Would a reader enjoy it? With some creativity and a bit of help from NLP, you can prompt better AI image prompts for your projects. Try out Jasper AI here and spark your imagination!
Depending on when you read this article, you may be in time to take advantage of the Jasper AI Black Friday sale, which includes unlimited AI Art generations for free (for 3 team members!) when you subscribe to the content writing tool:
Ready to join Medium?
Get unlimited access to Medium with my referral link, and you’ll also be supporting my ongoing writing at no extra cost to you. I’d appreciate it.
Who is Jim the AI Whisperer?
Jim the AI Whisperer offers advanced training in how to use AI generators to create stunning visuals, as well as to write original, compelling content using AI. If you’re interested in learning more, you can contact me here.
