Learning DALL-E: A Beginner’s Guide to Visual AI, Inspired by Tech Threads

With a shout out to the Tech Threads Community, especially you, Eleonor. Rose

Recently, my journey with generative AI skills and interests has expanded. Inspired by Eleonor Rose’s modern cartoon avatars, I experimented with visual AI, using guided prompts and platforms like GPT/DALL-E and BING/DALL-E. This post is a guide to what I learned, sharing prompts, and aimed at helping other beginners get started creating AI-generated images.

AI Platforms I Explored

• ChatGPT: Access to a GPT-4 integrated with DALL-E (paid account).
• BING: Bing chat with GPT-4 or Bing Chat Enterprise account.

My Experiments with Prompts

Prompt 1: The Modern Cartoon Avatar

Inspired by Eleonor Rose’s viral Threads meme, I used her template to create a 3D animated character. The outcome was an image of an over-60 female character, supposed to reflect a blend of modern animation aesthetics and personal traits.

Elenor.Rose’s wonderful series of avatar images

Elenor.Rose’s first prompt(how she made what she made): “Create an image of a stylized 3D animated character with a striking resemblance to modern animation movies. The character is a [Age/ Age Goup][Ethnicity] stylish [Gender] with [hair type, hair color, hairstyle & optional hair accessory], and [eye description & any unique face description]. [She/He/They] is [height descriptor, & build and unique body descriptors]. [She/He/They] wears a [outfit, accessories, and shoe description]. [Here you can add any fun option like a pet sitting in your lap or you holding an item etc.] [She/He/They] sits casually with a friendly demeanor on a giant, iconic [color] ‘social media’ logo cube that says “[social media]” that is central in the composition. The background has soft lighting with a slight vignette to draw focus to the character. There is a subtle ambient glow that suggests digital connectivity, and the character’s profile picture appears on the upper part of the composition, within a stylized social media interface website displayed in the back. [Her/His/Their] name “[Name]” is prominently displayed.”

In addition to this prompt, she shared a wonderful list of additional prompts on another page she built, check them out. There are also more instructions on a super-helpful Notion page. She also built a Social Media Avatar(GPT-4 needed for access) that can shortcut this process for you. Elenor. Rose’s Visual Prompt GPT, which helped me get language to describe how I look, is also super useful(need ChatGPT-4 to access it).

What happened when I tried it

This was my first effort.

I asked DALL-E to make the character look more mature and got this. FAIL

This was my fourth effort and the one I liked the most.

My version of the prompt: “Create an image of a stylized 3D animated character with a striking resemblance to modern animation movies. The character is an over-60 but fit and attractive female with short striking white hair, cut longer on top and close on the sides, almond-shaped grey eyes, pale skin, full dark pink lips, and glasses. She is short and curvy, with a slightly husky frame, fit-looking and well-proportioned. She wears a black ribbed turtleneck, light blue jeans, white socks, and black Brooks sneakers. She sits casually with a friendly demeanor on a garden bench with books by her side. The space should have a modern and digital aesthetic. The background has soft lighting with a slight vignette to draw focus to the character. A subtle ambient glow suggests digital connectivity, and the character’s profile picture on a laptop appears on the upper part of the composition, within a stylized social media interface website in the back. Her name “Susan Mernit ” is prominently displayed.”

Prompt 2: David Leal’s Doodle Challenge

David’s prompt: “Create a doodle of a white man in his mid-30s, with shoulder-long dark wavy hair, short stubble goatee, and simple eyes. He has a hopeful expression. He holds a coffee mug, waves hello, and wears a hoodie with an anchor logo, jogger pants, and flip-flops. He sits on a tall wood stool over an empty white background. The illustration is textured with noise, has coarse edges, and soft contrasts.”

What happened when I tried it

First effort; made a couple of others as well.

My version of the prompt: “Create a doodle of a white woman in her sixties, with a very short pixie cut of platinum blonde, almost white hair, and almond-shaped grey eyes. She has a joyful expression. She is holding a black cell phone and smiling at the screen with a reflected image of a young blonde boy, her 2-year-old grandson. , The woman wears a black turtleneck sweater and light-washed blue jeans, with white socks and black Brooks running shoes. She sits on a tall wood stool over an empty white background. The illustration is textured with noise, has coarse edges, and soft contrasts.”

Prompt 3: Zoom Chat Illustration

For my Quick Guide to Using the Zoom AI Companion. I tried to craft a prompt featuring the Susan Mernit avatar in a cozy, mid-century modern home office. It didn’t work as planned, but the Zoom screen was badass. Two versions:

My prompt: “Using Dalle-E, please generate an illustration I can use at the top of this guide. The main character should be the last Susan Mernit avatar that you created, portrayed in a cozy, midcentury modern home office, in front of a laptop and a large computer screen where a Zoom meeting with 9 participants, ranging in age from 25 to 70, with one an East Asian woman, two Black women, two bi-racial Asian men, a white man, two female Latinas, and one male Filipino.”

Prompt 4: A Pre-Raphaelite Twist

Working with an uploaded image: My prompt (after I attached a JPG of the painting to my chatbox): “This is an image of Janey Morris, a Pre-Raphaelite model in England in 1871. I would like you to portray her inside a library, sitting in a chair, pensively holding a book, with books and art objects around her, and a small white dog at her feet. there is a large paned window behind her, and it is late at night. The stars are visible and the moon is streaming in through the window. At the edge of the frame of the image, the viewer can see a large computer screen and a laptop, reflecting this same image in miniature, raising the question of what era it is.”

After the first image came back, I told the AI: “Please keep the same scene but put it inside more of a castle,” just to see what would happen. I got this:

Prompt 5: Illustration for Announcing AI Workshops

To announce new workshops, I visualized a scene of four women engaged in a Zoom session and shared it with the AI. The resulting images, while not entirely fulfilling my request, showcased a progression in my understanding and application of visual AI prompts that I feel good about.

My prompt: “Portray four women sitting at computers in a comfortable midcentury modern living room where there are tubs of plants, bookcases with books, and mid-century modern art hung on the walls. There is a large window to one side, and natural light is coming in. The four women are sitting at their desks, which are arranged in a semi-circle. They each have a laptop open. One woman is about 40, has shoulder-length blonde hair, and is wearing a light pink sweater, blue jeans, and white Keds sneakers. Another woman is thin, with short, straight brunette hair, and is wearing a black v-neck sweater and black jeans with black hiking boots. A third woman is short and slightly chubby, with cropped white hair, wearing a black turtleneck, light-washed blue jeans, and black sneakers. The fourth woman has shoulder-length curly brown hair and is wearing an olive green top and dark jeans. Together, they are facing a large computer monitor on a wall, where a Zoom session is happening. The screen for the Zoom meeting shows 9 participants, ranging in age from 25 to 70, with one East Asian woman, two Black women, two bi-racial Asian men, a white man, two female Latinas, and one male Filipino. The feeling of the illustration is professional and detailed, but the mood is positive and engaging.”

I kept asking the AI to tweak the image and then got the ones below. The fact that in every image all the women appear to be under 40, and none of them have short hair or white hair as reflected, really annoys me. I feel that this is a reflection of the DALL-E platform’s training bias, e.g. not enough diversity in the images and directions. I made several more versions to try to get better representation of diverse people, but the AI was not helpful. (That is infuriating. And more on this to come.)

Reflections and Next Steps I’m just starting to learn how to create with visual AI. There is so much to learn, and the tools are evolving so quickly.

But without the fun, generosity, and camaraderie of the #Tech Threads community, I don’t know if I would have gone over to the visual side of the AI aisle and checked it out at all.

The #Techthreads community is on Threads.net; come find me there at @susanmernit, but more importantly, connect with this wonderful group.

Note this post also appeared on my blog; find more AI articles there.