Gen AI hallucinations and predictable outputs driving you mad? Be inspired by Elvis Presley’s wisdom and use AI image generators like a pro
Generative AI backlashes, hallucinations, and predictable outputs driving you up the wall? “All this aggravation ain’t satisfactioning me,” crooned Elvis Presley in “A Little Less Conversation,” and we couldn’t agree more. But don’t throw in the towel just yet!
If your goal is to create truly captivating artwork, simply letting the AI beast run wild while you sit back won’t cut it. Instead, inject a bit of human creativity and mastery of the technical nuances, and voilà—magic happens.
Here are some tips to get the most creative output from the latest text-to-image Generative AI tools like Firefly, Midjourney, and DALL-E.
A Little Less Conversation, a Little More Action
Prompt design is a vital skill to master when using generative AI tools from the get-go. Two years ago, it felt like pure wizardry to type your “/imagine prompt” into Discord and watch complex visuals materialise on the screen. Now, the key skill for beginners is mastering the default parameters and options offered by AI platforms.
When working on augmented versions of an existing reference (say, a photograph you took or a character you sketched), it’s crucial to understand which parameters will enhance your project and which might flatten or ruin your creative input.
View this post on Instagram
A Little Less Fight, a Little More Spark
Battling with the AI to achieve a precise result is a futile, time-consuming effort. The wow factor often lies in the unexpected.
Using the latest versions of each text-to-image tool ensures better accuracy: thanks to their training on billions of parameters and reliance on larger LLMs, they understand more detail than previous systems. You can count on precise results and then tweak their core structure with your creative flair.
Try finalising complex visuals sourced from different media: start with a real picture, then add or adjust details, expand margins, and mix visuals. Give it a good stir, and enjoy your creative shake!
View this post on Instagram
A Little Less Bark, a Little More Bite
How can AI-generated artwork be made more impactful and creative while avoiding predictability? Excellent question. Most Gen AI tools are statistical models based on predictability, aiming to interpret text inputs with the closest representation.
But sometimes precision isn’t the goal; instead, focus on creativity and unpredictability. To achieve unexpected outputs, use very short or nonsensical prompts, including a few disconnected words or no syntax. Prepare to be amazed!
View this post on Instagram
The Ultimate Pro Tips You Can’t Miss
- Don’t procrastinate, don’t articulate! (cit. Elvis Presley): The more vague and less descriptive your prompt, the more unpredictable the result.
- Use the #chaos, #stylize, #weird, and #random parameters on Midjourney to explore unusual aesthetics or control the strength of default styles. #repeat and #seed on Midjourney are great for incremental work on a visual.
- Most text-to-image AI tools allow you to use a character reference to start a storyboard or create multiple images with slight variations. Use the “n” parameter on DALL-E to generate up to 10 images.
View this post on Instagram
- Use “insert/remove” and “add/subtract” to play with elements in the visual. Use the “--no” parameter on Midjourney to specify concepts you don’t want to include, which will help you erase unwanted pieces.
- Generative Fill settings on Firefly are fantastic: try the Generative recolor for fashion or artistic projects to speed up your process while experimenting with different colour palettes.
- Pick your favourite version! The default is the latest on every platform, but you can add previous versions as parameters on Midjourney (e.g., “--v 4”) and DALL-E (e.g., “dall-e-2”) for a nostalgic touch.
View this post on Instagram
Here, you can check out all the new parameters of DALL-E 3, Midjourney, and Firefly. Have fun!