AI Workflows for Art Deco Revival: An Architecture Design Guide

The AI Design Guide for Art Deco Architecture and Beyond

Introduction

Speed. Verticality. Opulence. Glamor. No this isn’t the roaring 1920s, it’s the roaring 2020s and the ongoing Art Deco revival the architecture world is experiencing. In this article, we’ll be investigating how to use AI workflows for design and architecture, how you as a designer can better understand this ongoing Art Deco revival trend, and how you can augment your architectural design process.

Outline

1. History
   1. Origins
   2. Influences
   3. Decline and Successors
   4. Modern Revival
2. How to Identify Art Deco
3. AI Workflows
   1. Prompt Crafting
      • Prompting 101
      • Prompt Adjustment
      • Image Prompting
      • Image Prompting and Realism
      • Design Evolution Techniques
      • Blending and Developing Styles
   2. Product-Inspired Design
   3. Pan-PromeAI Pipeline
   4. Preview: Interior Design
4. An Art Deco Revival: Should We Though?
5. Coming Up…

1. History

Origins

When trying to achieve a specific style with AI image generators, it is important to understand that style’s history and learn to describe examples before developing an AI workflow to explore your ideas.

Art Deco burst onto the world stage in the 1920s as a way to celebrate the widespread optimism that rapidly changing technology could bring about a better future, while offering some escapism after the horrors of WWI [1][2]. It quickly spread around the world as maximalist answer to the question “what does modernity look like?” and an alternative design philosophy to the minimalist, Bauhaus-influenced International Style seeking to answer that same question [3].   

Art Deco and Art Nouveau share the goal of adding beauty to everyday objects and architecture. However, in stark contrast to Art Nouveau’s organic shapes and sinuous curves, Art Deco revels in geometric patterns and sleek machined decoration. It relies on color, flat design, and simple geometry to allow the materials to tell their story. While art nouveau emphasizes the uniqueness in handmade objects, Art Deco takes a more egalitarian approach by celebrating striking designs that can be mass produced as a way of making a luxurious-feeling style available to everyone [3].

Influences

Initially Art Deco was developed as a design style for manufactured products and thus its motifs were often a direct result of 1920s manufacturing methods, not just a stylistic choice. Examples include parallel radiator fin lines (used to stabilize early Bakelite plastics in their molds), the ubiquity metallic finishes (due to newly commercialized chrome dipping and aluminum manufacturing), and the “in-motion” forms made economical by automotive manufacturing [4].     

Art Deco architecture incorporated this manufactured aesthetic among other influences. New thoughts about urbanism had a significant impact as well. For example, setbacks became a zoning requirement in Manhattan to ensure sunlight reached the streets in the skyscraper canyon below [5]. This incentivized a ziggurat form that become an iconic part of any Art Deco skyline. Another influence was a growing middle-class fascination with travel, so there were many works inspired by trains, airships, and ocean liners, as well as the destinations they could take you. Finally socioeconomic trends like the explosion of consumerism in the 1920s or a desire to market oneself as wealthy or sophisticated (as personified in The Great Gatsby), all fed into Art Deco’s lavish aesthetic.

Image References [i – vi]

Decline and Successors

Art Deco’s popularity persisted into the 30s despite the great depression. However, it began transforming into a new style, Streamline Moderne, which eschewed extravagance and ornamentation. It traded in jagged machined edges for simple teardrop-shaped forms based on the developing field of aerodynamics. It replaced the imposing verticality of Art Deco skyscrapers with more practical low-slung buildings emphasizing effortless motion and horizontality [6]. Further styles would evolve from Streamline Moderne such as Googie style, Mid-Century Modern, and a particularly colorful, postmodern, and tongue-in-cheek Art Deco revival movement in the 1980s thanks to designers like the Memphis Group [3][7].

Modern Revival

Another Art Deco revival has been quickly gaining traction since at least the mid-2010s and once again Art Deco is having its moment. Neutral is out and bold maximalism is in [8]. The Art Deco movement today borrows from the Gatsby-like opulence of the 1920s and takes a few cues from the neon art deco of the 1980s, but tends to execute them in a quieter, more reserved way to align with 21^st century tastes. Incorporating this style into your exteriors or interiors is now a way for designers to communicate one key concept: unique elegance.

So far most examples in façade design have been in high-end luxury residential developments like Rose Hill, Brooklyn Tower, and The Fitzroy in NYC or Pine Avenue in Melbourne [9]. Some other slightly less upscale developments exist (e.g. The Green House, Long Island City) but its most widespread adoption is in interior design, where a subtle more accessible art deco style noted above is gaining popularity.

[vii – xiv]

2. How to Identify Art Deco

Knowing how to describe specific Art Deco characteristics when crafting prompts is key to controlling in the AI generation process and is obviously important for traditional design methods as well. To that end, key features of Art Deco to keep in mind include:

[xv][xvi]
[1] [6] [10] [11]

3. AI Workflows

Pixels to Plans is about exploring AI workflows for design. So now that we have established a design vocabulary to prompt for (Art Deco), we will cover the most effective basic prompting methods for developing architectural concepts (in any style), as well as three more advanced methods uniquely suited for Art Deco design.

Prompt Crafting

Prompting 101

Whether they’re images, text, or parameters, prompts are at the core of all AI design processes. “Prompt engineering” has been a major buzzword this year, which I like to define as the systemic research and analysis of what makes a good prompt for a given task. However, “prompt crafting” in my opinion is the use of one’s creative intuition to take a prompt and evolve it in a more organic, artistic process.

The following AI workflows use Midjourney to generate images, but some of the concepts will apply to other AI image platforms. This article assumes a basic understanding of prompting. If this is your first time with AI image generators, click here [12] (Midjourney 101). If you want more of a deep dive on some of the newer features we’ll be exploring not specific to architecture, click here [13] (for image chain prompting), here [14] (for remixing), here [15] (for zoom, vary, shorten, and more), and here [16] (for the MidLibrary catalog of Genres and Art Movement prompts)

To start developing your own AI-generated concepts, it’s best to start with something simple. Let’s go with “art deco architecture” which gives us the first result shown below. As you can see, Midjourney training data must include a lot of examples of movie theater marquee towers and the ubiquitous plaster-clad Art Deco buildings Miami Beach. This results in a bias for Miami-inspired pastel plaster results like these. To be honest these are a bit bland with limited variety between images. The results simply struggle to capture the breadth of aesthetics Art Deco offers unless prompted in very specific ways.

Prompt Adjustment

How can we fix Midjourney’s Miami movie theater problem and produce something more original and striking? For this situation, we could consider the following options:

• Adding Keywords: Try additional keywords like unique, original, or the specific finishes you want your result to include
• Negative prompts: Use a —no  parameter to prompt against keywords like bland, boring, unoriginal, stucco, Miami, theater, or pastel if you want to get away from this aforementioned bias.
• Stylize: Increase the  —s parameter above 100 to incentivize more visually appealing and exciting images, though this may come at the cost of accuracy to the prompt
• Chaos: increase the —c parameter above 0 to incentivize visual differences between the four images each prompt produces. This too may reduce accuracy to the prompt for some of the images. 

In Midjourney, I’ve found the most reliable way to get more interesting results is increasing both chaos to between 10 and 50 (to disincentivize getting four similar compositions) and stylization to between 200 and 500 (to promote more creative prompt interpretation and more aesthetic results) as shown below. Does this method produce more Streamline-Moderne-inspired results? Sort of, but Streamline Moderne is just evolved Art Deco so I’ll give it a pass.

Image Prompting

Let’s take some of these results and try to develop their concept further. Since a picture is worth 1000 words, image prompting (i.e., uploading one or more photos as a reference for the AI) is by far the most powerful way to control the output of this process. In fact each result can recursively be fed back into the model (image chain prompting) to gradually dial in the results to be more consistent with much more control than words can offer.

The images in the next slide show were each made using combinations of the reference images below. In some cases, the previously generated images in the collection were used as part of the prompt as well. If you look closely, you can see the visual DNA of the reference images forming not just a prompt chain, but a web or family tree of interconnected ideas. Can you guess which images were used to to make each result?

REFERENCE IMAGES

Image Prompting and Realism

Say your model isn’t producing the photorealism or materiality you were hoping for. In that case, image prompting with close-up photos of the materials you want can help guide the model towards more grounded and photoreal results than simply describing them– a worthwhile trick for AI models that lean towards a more illustrated or rendered appearance.

As an example, here is an attempt to image prompt for the deep blue tile and metallic plaster/terra cotta cladding expressed on two-story Floral Depot building in Oakland, CA, but reimagined as a taller modern Art Deco-inspired development.

Design Evolution Techniques

These are all very good but say we want to emphasize practicality, with subtle yet distinct Art Deco flourishes paired with the everyday forms and materials an architect might consider for a multifamily development. Midjourney offers many methods of recursively evolving a design concept without starting from scratch on a new generation. To develop a concept further, two of the most reliable paths I have found are:

1. Image chain prompt using the best example of the previous generation as the new image prompt but with other images or phrases to tone down (or tone up) the craziness.
2. Use Midjourney’s “Vary Strong” feature to to produce similar images to a previous result that copy that result’s style rather than its composition. This lets you easily evolve a concept through multiple generations while strongly adhering to the original style.

Both of these methods, or a combination of the two, can work better depending on the context and it is up to the prompt crafter to determine which is best for their application.

Here are some results of these methods when used to evolve a group of practical art deco façade ideas that might be suitable for a multi-family apartment building:

Blending and Developing Styles

What if we identify an aesthetic you like and want to push it to its limits, reimagining it in new contexts with bolder interpretations, or blended with other styles to create a truly never-before-seen design? As an example, the following images show experiments in blending popular Japanese/Scandinavian design concepts (a.k.a. Japandi) with Art Deco, plus a burst of vibrant color. Using variations and image prompts you can expand one image into an entire collection exploring an aesthetic.

This image series aims to define an Art Deco-Japandi aesthetic that emphasizes subtle flowing lines, vibrant jewel tone blues, rich wood tones with vertical slats, general openness, and overall playfulness all with the subtle patterns and verticality of Art Deco in its DNA. Prompts, image references, and an abbreviated description of the design goal are included to help readers understand the process and experiment with similar methods.

Hopefully this demonstrates that with an AI architecture workflow as your design tool, there’s no excuse for not thinking outside the box.

Product-Inspired Design

As noted above, Art Deco architecture was often inspired by Art Deco products, taking the motifs of a car, a radio, or a watch and blowing them up to towering proportions. Similarly, generating a product concept in Midjourney can serve as the nucleus for a much larger design concept, and is a fitting design approach given Art Deco’s history.

For those detail oriented designers out there — those who like to think about day-to-day objects and centerpiece features first, then expand to envision the bigger picture — this approach can supercharge your design process.

Pan-PromeAI Pipeline

One of Midjourney’s drawbacks is the lack of control it gives you over the composition of an image. This is problematic in architecture where you always start with a site or massing and then need to develop it over countless iterations. Unfortunately with Midjourney and many other image generators, the output will usually be a completely different building than the input, where the only thing they share is their general vibe. Here are some examples of attempting to further develop a rough floorplate stack sketch over a Google Earth screenshot using Midjourney Version 4, that didn’t go according to plan, but at least made some interesting results.

The colors and style are sort of there, but the composition and massing is all over the place. Thus, to develop a design further we need to get creative with our workflows.

Once again, we start with our favorite scalloped window, but we would like to use it as the inspiration for a mid-rise tower. From there, we can use Midjourney’s pan function to direct the AI to start outpainting in a specific direction. This gives us more control of the resulting composition than a zoom would. We can repeat this process, tweaking the prompt as we go to first ask for the base of the tower, then the middle, then the top as we continue up the building. The key for each step is specifying the components associated with the part you want to generate such as “street, base” in the first part, “curtain wall, mid-section” in the next few, and “roof, crown, sky, top” in the final pan. The result?

Ok how about another panning project. This time we want a taller, more imposing, and stereotypically Art Deco skyscraper. We start with a large sculptural façade element over a door and imagine it in the context of our skyscraper, resulting in the following:

At a certain point perspective and aspect ratio may become an issue. You will be left with a long image that has lots of detail and structure, but generally has lower quality and poor realism due to the perspective issues.

That’s where sketch-rendering AI comes in to do what Midjourney cannot. In this example we will look at PromeAI, a freemium AI platform with plenty of architecture-focused tools that provides sketch-to-image capabilities and other similar features. By using Midjourney outputs in PromeAI, we can take advantage of the relative strengths and weaknesses of each platform. Specifically, PromeAI’s Creative Fusion feature lets the user blend the composition of one image with the style of another, offering much more control than Midjourney’s blend feature under certain circumstances.

Now with a rough sketch to define our massing and our long Midjourney image to define articulation at the base, middle, and top of the tower, we can envision how they might work together.

Imagine the freedom an AI workflow like this this could give to a design firm. It would allow one designer to develop massing and another designer to develop façade articulation in parallel completely independently of one another. Then all they need to do is periodically fuse their outputs together to develop insight into which ones might work well together in a matter of seconds.

Now you may also ask whether that whole initial panning step is necessary. While not always necessary, it is important to pair images that are of a similar scale and subject matter. If you pair an overview sketch with a close up image like our original window, we may get a more surreal close-up result as the AI struggles to combine the two concepts. These may produce their own inspiring results like the imposing lacquered finishes it settled on in this example below, but that all depends on selecting the right inputs for what you as designer are trying to accomplish.

Preview: Interiors

As with Art Nouveau, Art Deco interior design deserves its own post. However in the meantime I’ll leave you with a brief fourth workflow idea for art deco design (or any interior design for that matter): The prompt “award winning photo of [style] interior design, [style] interior architecture mood board collage” is great for finding interior design inspiration:

“History isn’t linear. It’s cyclical. What was once avant garde becomes mainstream, then outdated, only to be rediscovered and appreciated again or for new reasons.” – Stewart Hicks [16].

Art Deco design may be on the verge of coming back in full force but I’ll admit, this Gatsby-like style can come off as a bit ostentatious at times. It definitely retains a timeless appeal at its core but the key challenge for designers today will be finding ways up update it for a modern audience.

Will this revival follow the path of the neon-drenched 1980s Art Deco pastiche [17][18]? It would certainly be fun if it did. Alternately will it stick to its roots as a serious way of expressing grandeur and affluence seen only in condos and office towers? Will it take a tactful middle road and aim to be a subtle celebration of timeless design concepts targeting the widest possible audience? Or perhaps Art Deco doesn’t take off and some new style seizes the zeitgeist. Either way, we’ll certainly be exploring many other contenders on this blog, so that’s not out of the question.

Some may disagree with Art Deco. Some may love it. Some may look at this whole blog and say “The 1900s called. They want their architecture back”. This blog is not meant to be my way or the highway. It is a series of tableaus to peruse through and find something inspiring, whatever that may be. It’s a stream of ideas paired with practical design tips, tricks, and workflows, with the ultimate goal of helping designers navigate the world of AI design as they bring creative new architectural concepts into reality.

Some of the examples above touched on leveraging the strengths of two different AI tools to compliment each other. In the next post, we will compare some of the best AI image generators for architecture and how to combine them in your design workflow.

---

---

Newsletter Sign-up

---

Citations:

[1] YouTube – Wayfair – The Rise, Fall, and Revival of Art Deco | A Style Is Born W/ @KazRowe

[2] New York Public Library – Art Deco: A research Guide

[3] The Art Story – Art Deco

[4] YouTube – History of ID Matthew Bird, RISD – Art Deco

[5] My Modern Met – Art Deco

[6] YouTube – Michael Pacitti – Art Deco vs Streamline Moderne 4K

[7] Yesterday’s News -Art Deco Revival in 2020

[8] The Spruce – This Retro Design Style Is 2023’s Next Biggest Trend

[9] Dezeen – Art Deco Architecture Design Projects

[10] YouTube – Nick Lewis – 4 easy tips to bring art deco into your home

[11] Coaster Furniture – Art Deco furniture 101: How to add glitz and glamor to your home

[12] YouTube – Wade McMaster – Creator Impact –  MIDJOURNEY Tutorial: How to get started in 2023

[13] YouTube – Future Tech Pilot – This Midjourney Technique Will Put You in the Top 1% | Chain Prompting Tutorial

[14] YouTube – Future Tech Pilot – The BEST Thing About Midjourney Version 5.2

[15] YouTube – Theoretically Media – Mastering Midjourney v5.2 : A Deep Dive into the New Version! (Zoom, Vary, Shorten, Tips & Tricks!)

[16] MidLibrary – Architectural Genres

[17] YouTube – Stewart Hicks – The 1980s Design Secret We’ve Forgotten

[18] Instagram – Meg Gustafson – 80s Deco

Image References

[i] https://www.linearity.io/blog/art-deco/

[ii] https://www.timeout.com/los-angeles/things-to-do/art-deco-buildings-los-angeles

[iii] https://archi-monarch.com/art-deco-design-in-architecture/

[iv] https://unsplash.com/photos/a-yellow-and-white-building-with-a-blue-sky-in-the-background-tqss4wZILRQ

[v] https://matadornetwork.com/read/coolest-art-deco-buildings-miami/

[vi] https://www.miamiandbeaches.com/things-to-do/history-and-heritage/art-deco-historic-district#:~:text=The%20first%2020th%2Dcentury%20neighborhood,built%20between%201923%20and%201943.

[vii] https://interiordesign.net/awards/nycxdesign/2023/cetraruddy-architecture-rose-hill/   

[viii]  https://www.skyscrapercity.com/threads/the-brooklyn-tower-new-york-325-m-74-fl.2400486/

[ix] https://www.woodsbagot.com/projects/the-brooklyn-tower/ 

[x] https://www.behance.net/gallery/77819671/180E-88th-St-Penthouse

[xi]  https://www.pineave.com.au/architecture.html

[xii] https://www.bloomberg.com/news/features/2023-03-18/a-new-old-art-deco-tower-marks-manhattan-s-terra-cotta-comeback

[xiii] https://www.dezeen.com/2019/06/07/art-deco-architecture-design-projects/

[xiv] https://www.archpaper.com/2023/09/studio-v-the-green-house-terra-cotta-cladding-long-island-city/

[xv] https://mymodernmet.com/art-deco-architecture/

[xvi] https://www.theartstory.org/movement/art-deco/