How to design like [you name it]
Joan Fusté
Software Development Veteran | Backend Specialist | Current Master’s in Big Data | AI Magazine Innovator | Pixed Magazine Editor | AI-Imaginator
.. and not to mention her.
I’ve spoken about this topic on a few occasions: how to use the style of an artist, photographer, or creative without mentioning their name in the prompt of a generative AI, any AI.
It’s not always easy, but in some cases, it gets pretty close to what we’re looking for. In this case, I’ll use MidJourney, which is my preferred tool.
How do we create portraits in the style of Steve McCurry without mentioning him?
Or anyone else, like a painter (e.g., Picasso) or an architect (e.g., Gaudí). First, let me say that the main reason is to avoid copyright conflicts, etc. But the second reason, and the most fun, is to create a simple yet sufficient description to recreate their style—and maybe even improve it.
How do we do it?
The first step is to use ChatGPT to generate a description of the style. After reading Geoffrey Packer post and his “Tim Burton-style” illustrations without using his name, I'll use Tim Burton for this example. As soon as I read it, I remembered that some time ago, I recreated various styles using the following technique.
A simple prompt asks ChatGPT to describe the style of [person] in 25 words, separated by commas, and in English (the language is specified here because English is not my native language). In the case of Tim Burton, ChatGPT responded with:
Gothic, whimsical, eerie, surreal, dark, quirky, dramatic, macabre, mysterious, shadowy, curly, vintage, fantastical, moody, theatrical, eccentric, creepy, melancholic, textured, imaginative, stylized.
It’s pretty accurate, but it’s missing some key elements. Without these, any image using this stylistic description might appear very gothic, sure, but not necessarily close to the desired style.
So, we refine ChatGPT’s prompt to include his typical curls, and for more detail, we provide a final image of what we’re looking for. In this case, ChatGPT returns the following style:
A gothic, surreal architectural style with dramatic curves, spiraling elements resembling curls, intricate stained glass patterns, muted earthy tones, and an eerie yet whimsical atmosphere. The design combines organic, flowing forms with sharp spires and textured, cracked surfaces, evoking mystery, fantasy, and a dreamlike quality.
It’s quite comprehensive, but let’s use it and see what happens. The prompt is very simple: I just added the word “render” before the previous description.
Amazing! It’s almost perfect, and I’d even say it’s better than what I was looking for.
I’m not a fan of such a long description, so now we have a fantastic option: moodboards. Using a 2x2 grid, I create a moodboard to obtain the code --p m7292861484164448273 and use it to create a preference like this:
/prefer option set TimBurton value: --p m7292861484164448273
Now, I can create much shorter prompts with the same style, adjustable with more images (one of the wonders of moodboards), and achieve this:
{render,Lambo} --TimBurton --ar 16:9
What do you think? We now have “Tim Burton-style” images without using his name in the prompt. The prompts generated from the above are:
render --p fwvxqxr --ar 16:9
Lambo --p fwvxqxr --ar 16:9
So, we’ve achieved the goal.
Apply the technique to any artist, refine it a bit, create a moodboard, refine it further, and there you have it.
Let me know if you need some help!
Co-Founder @TheCuriosityCo-op | CNA Turned AI Artist | Making Technology Human Through Visual Storytelling | Midjourney, ChatGPT & Claude Explorer
6 天前The workflow makes it simple to understand and accessible to anyone with access to an LLM and image generator; though, in this case, Midjourney is the better one to use. ????