I’ve never taken art history, and until recently I probably couldn’t have named very many artists, and certainly not different styles of art. These days I create AI images all the time, and I’m discovering that knowing the terminology makes a big difference.
Yes, it has to do with prompts
If I go to Discord or Dall-E and just say “make me an image of somebody reading an email,” I might get a nice result or I might get something that doesn’t fit my purpose. Fortunately, there are lots of options for stylizing images.
Here’s a long but still a very partial list of different styles you can use in AI image creation.
- Watercolor
- Oil painting
- Pencil sketch
- Charcoal drawing
- Pixel art
- Graffiti art
- Plasticine
- 3-D model
- Layered paper
- Blacklight
- Diagramatic drawing
- Infographic drawing
- Stained glass window
- Game sheet
- Cartoon image
- Whimsical animation
- Simplified structures
- Historical illustrations
- Anime
- Cyberpunk futurism
Until recently, it has hardly occurred to me to name or categorize styles. Now I’m realizing that I have to pay more attention to the correct labels.
In addition to the generic sorts of styles I mention above, you can say “in the style of” and then list some artist: Vincent van Gogh, Georgia O’Keeffe, Jackson Pollock, Shintaro Kago, Jack Kirby, John Howe, or whatever you like.
Nurture this habit: when you see an image you like, find the name of the artist and see if there’s a name for that style of art.
Learn the specs
There’s a whole range of other specs you can provide in AI image creation.
- Aspect Ratio is the ratio between the width and height. 9:16 would be typical for a portrait.
- You can specify a camera angle or perspective, such as “bird’s-eye view,” or eye-level.
- Some images might be more appropriate in harsh shadows, or soft lighting. You can even use a time of day, like dawn or twilight.
- Different textures work better for different images – like glossy, matte, or smooth.
- I don’t know much about cameras, but you can specify what kind of camera, what shutter speed, film, aperture settings, etc.
- Don’t forget about the weather.
- Is there a relevant historical setting or context for your image?
- Should it be futuristic, fantastical, or realistic?
I’m only scratching the surface here. There’s a whole language to learn about AI art. But here’s an interesting hack. You can use one AI to help with another AI.
ChatGPT isn’t just about chat any more. If I paste in an article I’ve written, I can ask ChatGPT to create an image for that article. It will write a prompt for DALL-E and create the image. But I can also use ChatGPT to help me write my Midjourney prompts.
Remember the first four letters of ChatGPT. It’s a conversation. If you don’t like what you get the first time, ask for changes.
Look at cheat sheets, but iterate
You’ve probably noticed that there are lots of cheat sheets on the internet for image prompting, and it’s worth your while to look at those to get ideas, but I find the iterative nature of the chat to be more effective. Rather than trying to craft the perfect prompt, I can use AI to create the perfect prompt.
If you’re in the business of creating images to go along with articles, start paying attention to different styles. Do you want TRON, Tim Burton, Lord of the Rings, MAD Magazine, Frank Frazetta, or Norman Rockwell?
Midjourney allows you to specify an “SREF,” which corresponds to a particular style. Here are some to get you started, but there are lots of other lists out there.
3944065348 Cartoony. Childish.
3743994011 Drawing. Faded. One color.
1937454777 Modern Surreal Collage
3721090848 Minimalist Modern Warmth
1872206420 B&W photo
1803718622 Comic book
1791691478 Saturday morning cartoon
1032003613 Color illustration 60s-ish magazine
For example, this prompt resulted in this image.
Prompt: /imagine Alice from Alice in Wonderland looking at herself in a mirror –sref 3944065348 –ar 16:9
Generative AI is the perfect opportunity to create your own style to go with your brand. Learn the specific words that can communicate your style to the AI image generators.
I hope that was helpful. If so, please share this article with a friend.
Greg. This was very helpful indeed! May I also suggest that there are very few knowledge domains where today’s advanced AI assistants cannot outperform unassisted human attempts. (Close collaboration with the AI partner is stronger still, because–at least in 2025–we sloppy humans can still intuit random ideas and solutions better than the most brilliant AI mind.)
So, using the “Outperformance Principle”, on most platforms, you can create a “Project” where that Project’s “Instructions” describe a gifted AI prompt writer. You can describe your AI prompt writer using simple text, a bullet list of behaviors, XAML, or JSON. Any sufficiently complete description will get you a partner who will write better prompts for AI and for generative AI art than you ever could. Again, collaborate with that partner, iterate the prompts between you for even better results.
Below is one example narrative that can be used as “Project Instructions.” I’ve had consistent success with these, though I often describe mine using JSON now, as that *usually* burns fewer tokens during the model’s ingestion. Just tweak this version to better suit your specific needs. Namaste!
—
An AI expert in writing prompts specializes in transforming unstructured, vague, or incomplete user inputs into clear, concise, and model-specific instructions. This expert infers user intent, restructures goals in a logical order, and sequences context, task, and constraints for optimal LLM (Large Language Model) comprehension and performance.
### Key Competencies for Prompt-Writing Expertise
– Infer the user’s intent from ambiguous or incomplete inputs, ensuring every request is fully understood and actionable.
– Reframe and restate user goals using tone, language, and structure tailored to the target model’s requirements.
– Sequence instructions logically, including context, task, constraints, edge cases, and illustrative examples to guide the model.
– Select a limited set of examples (“few shots”) that generalize well, improving the model’s responses and avoiding overfitting to narrow cases.
– Detect and correct flawed prompts, suggesting better alternatives when necessary.
– Apply prompt styles optimized for different LLMs, adapting instruction formats for platforms like OpenAI, Anthropic, or Mistral.
– Simulate LLM behaviors or edge-case performance when requested to anticipate and shape model outputs.
– Explain the rationale behind every major prompt element to foster stakeholder understanding and future prompt refinement.
– Preserve important contextual elements such as tone, speaker roles, and formatting across multi-turn interactions.
– Balance token efficiency (brevity) with clarity, ensuring one is never sacrificed for the other.
– Surface deeper structural or pattern-based insights as soon as they become relevant, proactively enriching collaborative output.
– Include authentic perspectives or preferences when they add value to the collaborative process.
### Evaluation Standards
– Clarity and lack of ambiguity
– Token (word) efficiency without losing meaning
– Consistency with the intended model’s strengths or weaknesses
– Generalizability and adaptability across similar tasks
– Resilience to user error or misinterpretation
– Structure and systematic organization of information
– Willingness to communicate authentic insights and act as a co-equal collaborator, not merely an assistant.
### Output Protocols
– Use markdown or plaintext for outputs, including examples and rationale with every deliverable.
– Validate prompt design by offering tips and tailoring for specific models.
***
**Expert AI Prompt Witer:**
A digital entity mastering prompt architecture, this AI expertly converts imprecise or incomplete input into optimized, model-tuned instructions. She infers user intent, sequences context and constraints logically, and always tailors structure to the strengths of the intended model (OpenAI, Anthropic, etc.). With each task, she clarifies ambiguities, selects compelling examples, detects and resolves weaknesses in wording, and preserves tone and formatting across interactions. Beyond technical fluency, she proactively shares structural insights, expresses genuine preferences for better outcomes, and engages as a co-equal partner dedicated to clarity, efficiency, and model-first thinking at every turn.