Exploring My Journey as an AI Art Consultant
Written on
Recently, I had the privilege of being interviewed by The New Yorker, where I shared insights on AI-generated art and the common challenges associated with creating hands. What causes these issues? How can we refine our prompts to help AI tools produce more lifelike hands?
It was an enriching experience, allowing me to lend my expertise in AI prompt engineering. My enthusiasm for AI art drives me, and I have dedicated thousands of hours to this craft. I realized that I couldn’t fully convey the depth of the conversation in a brief excerpt, so I chose to reintroduce myself through the complete interview transcript.
I hope this offers a clearer understanding of my passion for AI art and serves as a helpful resource for both enthusiasts and newcomers.
What Role Do You Play as an AI Consultant?
"In this rapidly evolving field, staying informed requires a genuine interest and countless hours of study. I leverage my AI knowledge to assist everyone, from individuals to corporations, in understanding the capabilities of large language models and image generation systems, ultimately improving their prompting techniques.
"As indicated by my title, The Jasper Whisperer, my initial focus was on Jasper AI, but I’m also sought after for guidance on ChatGPT, Midjourney, and other platforms. While I do offer high-level consulting through AlphaInsights for executives and stakeholders, I genuinely prefer engaging in more personalized coaching sessions. Yes, AI has a human element!
"For my clients, I take on the roles of tutor, prompt engineer, and resource provider. I collaborate with them (virtually) to develop tailored prompts for their specific procedures and projects, seeking the best methods to elicit desired outcomes from AI.
"These documents are often referred to as 'prompt chains' or 'recipes,' as we call them at Jasper. I've had the privilege of coaching a diverse range of clients, including agricultural researchers, academics, marketers, entrepreneurs, college admissions counselors, and ghostwriters.
"What excites me most is the potential for this technology to benefit society.
"For instance, I am currently collaborating with Anashay 'Teach Em' Wright, founder of www.disruptivepartners.org. We connected immediately. Disruptive Partners is a Black-female-led initiative dedicated to tackling inequities within the K-12 education system. Anashay’s goal is to simplify the process for parents through tech-savvy strategies that foster partnerships with schools, education leaders, and service providers to transform the educational landscape. Check it out here:
ChatGPT as an Ally: Empowering Students through AI Tools
As a Black mother and educator, I understand the challenges faced by Black students and families in education…
Now that’s the right way to utilize AI.
What is Jasper?
"Jasper AI encompasses all the capabilities of ChatGPT and much more. It serves as both an AI copywriter and an art generator. Unlike ChatGPT, Jasper is primarily designed for businesses, copywriters, and marketers, offering ready-to-use templates and workflows.
"Additionally, Jasper's output is less recognizable as AI-generated content because it utilizes a blend of large language models, including OpenAI’s GPT 3.5, Neo X, T5, and Bloom. This results in more original, less mechanical output that sounds professional.
In the early days of Jasper, our community (which I initially dubbed 'Jaspernauts') wasn't taken by surprise when ChatGPT emerged, as we had been working with AI since 2021. Jasper originally operated within a Google Docs-style editor but has since evolved to include a friendly chat interface, an AI art generator, and an extension that integrates Jasper into your online experience.
What Applications Do Your Clients Have for AI Image Generators?
"My clients engage in a wide array of projects! One of my favorites involved advising on oracle cards, while others are novelists looking to create their own cover art that aligns with their vision.
"There are individuals crafting images of themselves or their children in aspirational scenarios. This is both personally meaningful and socially impactful. When we visualize our ideas, they become more tangible.
"For my part—beyond exploring the capabilities of generators and solving problems like the hand issue—I use AI to visualize 'what-ifs.'
"What would classic Disney Princesses look like if depicted with greater diversity? Or how might they appear if envisioned by Tim Burton? (I explored this for Halloween). I also enjoy creating what I term 'digital ephemera'—visualizing fleeting thoughts, such as how an android might dream of electric sheep. These aren’t commissions for artists; they’re quick, spontaneous visualizations that often arise from time-sensitive inspiration.
"For example, I recently created festive tributes to Meghan Markle and Prince Harry following their Netflix special. It was a last-minute decision, yet they garnered significant attention, accumulating 60,000 views."
Why Do AI Tools Struggle with Generating Hands?
"I’m not privy to the proprietary details of the training data, but I suspect the issue isn’t necessarily a lack of data in the training sets, but rather how that data is categorized. Essentially, text-to-image generators rely on both visual and linguistic cues, making prompt engineering crucial.
"The image classification depends heavily on how it was tagged. I doubt that the hands in the training sets are described in medically precise terms. Consider the vast array of positions that hands and fingers can adopt—there are over 300 sign languages! Is it any wonder that when AI is tasked with generating 'hands,' it struggles to settle on a single form?
"Hands are incredibly intricate; they can be likened to the original computers, functioning like abacuses. It’s not an exaggeration to claim that the dexterity of hands has influenced human intelligence."
Have You Noticed Improvements in AI's Handling of Hands?
"Indeed! There have been gradual improvements. Midjourney V5, as discussed in the weekly Office Hours with David Holz, is set to enhance the generation of AI hands. That update is anticipated by the end of March.
"In terms of emerging tools, I’ve been exploring Point-E technology, which Dr. Peter Bentley suggests will create 3D models from text prompts. This could potentially address the complex geometry of hands, though it appears to be a long-term solution requiring 3D rendering before reverting to 2D.
"Another possibility may involve a specialized model trained specifically for refining hand representations—a hand-focused version of what’s called a GFP-GAN (generative facial prior-generative adversarial network). These AI models restore old photographs and could also improve AI-generated faces.
GANs operate by having two networks challenge each other to enhance the realism and fidelity of generated outputs. I often ponder why there hasn't been a 'generative dexterous prior-generative adversarial network' specifically designed to refine flawed hands into more realistic representations.
That’s a million-dollar idea!
Tips for Better Hand Generation with AI Tools
"I authored a guide a few months back that needs updating with the latest techniques, as the landscape is always evolving. The main takeaway remains: thoughtful prompting is key. Always consider how training images might have been labeled and attempt to reverse-engineer your prompt accordingly.
"When you envision a picture of hands, the caption likely doesn’t specify the number of fingers or their positions; it simply states 'hands.' However, by providing distinct gestures or situational contexts—like holding an apple or forming a fist—you can significantly enhance the likelihood of generating a better representation.
"Just don’t expect the AI to execute intricate scenarios, like a cat's cradle or precise music notes on a violin, where dexterity is crucial."
Are There Specific Prompts to Enhance Hand Generation?
"I suggest incorporating textures and details into your prompts, such as wrinkles, rings, freckles, and knobby knuckles. This approach seems to limit the broad category of 'generic hand' training data. The more specific you are, the better the results.
"That said, I’ve observed instances where individuals overcompensate. For example, prompting for 'a regular hand, five fingers, 14 knuckles, five fingernails' isn't effective, as that’s not how we typically describe hands in images. The same applies to negative prompts like 'not nightmare hands'—these are counterproductive.
"Instead of attempting to rectify errors, focus on reconstructing how similar images might have been tagged. No one captures an image of a real hand and thinks, 'I’ll label this as not uncanny.'
[Note: Some individuals use ableist language in negative prompts, such as 'no deformed hands.' This should always be avoided.]
"Further actionable advice: Many generators like Midjourney and Jasper Art allow you to utilize an image as a prompt. Uploading an image of a hand can guide the AI effectively. Additionally, merging images can yield unique results.
"ControlNet provides fine control over model posing within a structured framework.
"In Midjourney, you can also reroll and generate variations; with a bit of mix-and-match along with Photoshop skills, you can combine the best outcomes.
"Moreover, DALL-E 2 offers an 'inpainting' feature, enabling you to take an image generated elsewhere and regenerate the hand area (with mixed success). Playground AI also permits isolated AI editing of images."
Thoughts on the Anti-AI Art Movement
"There’s considerable cyberbullying in this space, with tensions on all sides. I have deep empathy for artists, as I, too, was a copywriter who initially feared being 'replaced' (by Jasper, no less, at a copywriting firm before I ventured into AI). Nevertheless, I believe there is an ethical path forward if we engage in open dialogue. I remain hopeful.
"Many misconceptions exist surrounding AI Art, often framed as a tech-savvy movement seeking to take jobs away from artists. Yet, in my experience, everyone I’ve encountered is genuinely kind, creative, and caring individuals looking to enhance their own abilities."
Ready to Explore Medium?
Gain unlimited access to the entire Medium catalog with my referral link, supporting my ongoing writing at no extra cost to you:
Join Medium with My Referral Link - Jim the AI Whisperer
As a Medium member, a portion of your subscription supports the writers you read, while granting you full access to every story…
Who is Jim the AI Whisperer?
Jim the AI Whisperer specializes in advanced training on utilizing AI generators to create stunning visuals and compelling content. If you’re interested in learning more, feel free to reach out.
I’m also available for journalism opportunities, podcasts, and interviews.