The Best AI Image Generators for Architecture

Strengths, weaknesses, and workflows for combining AI tools in your design process.

Head-to-Head Comparison

Introduction

If you’re confused about which AI image platforms you should learn to integrate AI into your workflow, you’re not alone. A plethora of AI platforms have appeared on the market in 2023 and it can be hard to determine which is right for your practice. Well fortunately, we at Pixels to Plans did the legwork for you and analyzed 8 of the best AI image generators for architecture to see how they stack up. This article presents a comprehensive study of each platform, the tools they offer, and how to leverage the strengths of each one to develop an AI design workflow for your projects.

The Contenders

To start off, let’s look at the competition:

1. Midjourney – Arguably the gold standard of high-quality image generation in 2023. This is a high-creativity platform known for its coherency and aesthetics. Subscriptions start around $10/month as of November 2023. Its two main styles we will explore today include default (a more metaphorical, artistic style) and raw (a more literal, photographic style) but users can now develop custom styles to hone their personal aesthetic.
2. Playground AI – A freemium AI platform based on the open source model Stable Diffusion XL. It offers fast generation, a high daily image generation cap, one-click filters, powerful photo editing, inpainting/outpainting, and image-to-image controls as well as a new generation preview feature. This platform is great for both rapid ideation and refining ideas from other platforms.
3. PromeAI – This is another freemium platform that put sketch-to-render tools on the map. It offers a broad array of tools and has an extensive list of present styles specifically for architecture. Its free plan is quite limited though, so a subscription is necessary to get the most out of it.
4. Look.X – This is an AI platform specifically designed for architecture. It rose to prominence after being featured in an article by Zaha Hadid Architects alum and AI design proponent, Tim Fu. It focuses more on realism than creativity but offers a serious, grounded aesthetic many architects may want.
5. Dall-E 3 (Bing Image Creator) – This is the model used in the Bing Image Creator platform and offers fairly good artistic and text effects. It is developed by OpenAI, making it part of the ChatGPT/Microsoft/Bing ecosystem, and can be accessed directly by GPT 4.
6. Limewire AI Studio (Formerly BlueWillow) – No this isn’t a meme. Limewire is somehow back after buying AI image generation startup Blue Willow, one of the early free alternatives to Midjourney. Its capabilities are limited but it offers 10 free generations a day and has a slightly different aesthetic compared to other platforms that some users may like. In earlier versions it was excellent at tasks like cityscapes and elevations that Midjourney struggled with, but its current strengths for architecture have yet to be discovered.
7. Vizcom – This is largely an automotive and product-design-focused platform for sketch-to-render and creative fusion tools. It prioritizes giving designers more authorship compared to more randomized image generators and offers a high level of control over developing existing design concepts further. It can produce good results with architecture, but is only with the right workflow.
8. Veras – This is both a web app and plugin to Sketchup, Rhino, and Revit. It appears to offer no original image generation capabilities but offers a powerful model-to-render tool that lets you quickly use AI to generate photorealistic renders of your 3D models from any angle.

Prompt Benchmarks

We will be comparing these platforms using five prompt benchmarks (carefully selected prompts meant to test AI models). The concept of benchmarking is rooted in computer science and LLMs, but for AI art and architecture they are best exemplified and explained by Andrei Kovalev’s Midlibrary, a non-profit educational initiative that catalogues and blogs about about various Midjourney styles.

For this study I developed the following benchmarks:

1. Modern Art Nouveau
   • Prompt: Award winning art nouveau revival home architecture, modern sensibilities, exterior facade, stunning ornamentation
2. Rainy Usonian
   • Prompt: rainy usonian architecture, cozy bright warm lighting inside windows, exterior facade
3. Architecture
   • Prompt: architecture
4. Stylish Interiors
   • Prompt: Award winning photo of a stylish interior design, interior architecture mood board collage
5. Solarpunk Cityscape
   • Prompt: solarpunk cityscape photo, lots of buildings, panoramic, lush plants, dawn mist, civil twilight, intricate detail, hyperrealistic

I selected these prompts to showcase how each platform deals with different concepts. These five prompts assess how the platform balances intricate organic maximalism with subtlety and practicality, how it depicts minimalism as well as atmospheric effects, what it views as quintessential architecture, how it interprets the word stylish, and how it handles futurism.

Benchmark Results

The graphic below, and the enlarged version you can explore on Canva, show how each of the eight platforms and three multi-platform workflows handle these benchmarks. In the last four rows covering platforms that focus mainly on image-to-image generation methods like sketch-to-image and creative fusion (more on that below) the series demonstrates how two image-to-image platforms used together can produce phenomenal results.

You can zoom in on Canva to better understand he nuances of each output.

AI Image Generator Comparison by AcanthusAlchemist

Benchmark Takeaways

How well do you think each platform adhered to the prompt? Were there any results that surprised you? While I’m sure you can spot some trends yourself, some of my key takeaways from the study include:

• All models have vastly different interpretations of the word architecture
• Look.X, Veras, and Vizcom struggle with Art Nouveau unless you give them a photo example.
• Conversely, Dall-E 3 and Default Midjourney get a little too carried away with the Art Nouveau benchmark.
• Modern and minimal keywords like Usonian or stylish produce fairly consistent results across platforms.
• Midjourney, Playground and PromeAI put more fantasy into their Solarpunk futurism, while the others take a more grounded photographic approach.
• Limewire has a darker, moodier aesthetic
• Look.X and Vizom tend towards more muted color palettes. However, Vizcom’s new creative fusion tool  is a completely different story with much more potential (more on that below)

If you would like to see more studies like this or have benchmark/style ideas you would like this blog to explore, you can contact us with questions and suggestions. For notifications of new and related posts, you can subscribe in the sidebar at the top or bottom of the page.

Selecting The Best AI Image Generators for Your Project

Now that we understand the general vibe of each AI model, you need to decide which type of platform best fits your design goals. Most image generator tools can be broken down into the following categories:

1. High-Creativity (Low-Consistency) Tools
2. Constant Composition Tools
3. Sketch Rendering
4. Creative Fusion

High-Creativity (Low-Consistency)

This type of AI image generator needs no introduction. It is the type of tool covered in the last two posts [Post 1: Art Nouveau] [Post 2: Art Deco and Workflows] and throughout the news recently. For more information on how to leverage these tools, follow the links above.

In my opinion, the best AI image generators for high-creativity brainstorming early in design are Midjourney and Playground AI. Almost every AI image generator offers this capability (excluding Veras and similar 3D model focused plugins) but their quality, speed, and versatility varies. This type of tool is most impactful at the very start of a conceptual design phase. You can review the infographic above to determine which platform best fits the styles you want to explore or craft your own benchmarks for your use case.

A high-creativity model is excellent for brainstorming, but is terrible at maintaining consistent composition or massing between generations, meaning elaborating on a design concept can be challenging.

Constant Composition

Constant composition tools are a style of image-to-image that take a reference image, apply some noise or blur to it, and use that as the background noise instead of the white noise field Midjourney and other from-scratch high-creativity image generators rely on. This maintains the general composition of the reference image and lets you reimagine it. You can also vary how closely the result should match the base image.

A good example of a constant composition tool can be found in Playground AI. Playground allows for rapid variation, especially with its one-click “filter” feature, which lets you quickly append predefined prompt snippets to your base prompt. They can produce specific aesthetics like “Cinematic” for a cinematic effect, “RealViz XL” for realism, or my personal favorite…. “Woolitize” which turns everything into yarn (See slide 12 below).

The following flipbook provides a good example of constant composition using a futuristic solarpunk art nouveau veranda (which I originally made in Blue Willow and outpainted by Dall-E 2) as the base image. I used a variety of Playground’s filters to quickly reimagine the scene. You will notice this model retains but reimagines the focal points, key elements, and strong lines of the original image in new and surprising ways.

Playground Filter Flipbook by AcanthusAlchemist

With Playground’s filters and creative fusion, you can discover and analyze different styles to produce amazing results with minimal knowledge of prompting or art styles (but you might learn a few things along the way).

Sketch Rendering

Sketch rendering is another flavor of image-to-image that offers even more control. Already have a sketch or 3D model of a building but can’t be bothered to pick materials and sit around waiting for it to render? Try sketch rendering in platforms like Vizcom, Veras, or PromeAI.

Each of these took about 10 seconds. Starting to sound a lot better than your average render time isn’t it?

Creative Fusion

Let’s say you want to combine concepts from two buildings into a single image that uses the massing of one but the style of the other. If you combined these two in Midjourney, the result would be a completely different building than either of the two inputs as shown in the image prompting examples of a previous post.

If you attempted this in a constant composition or sketch rendering platform, you would quickly realize your options are limited to loosely controlling the composition of an image directly, while the style still needs to be described entirely with words.

Platforms with Creative Fusion offer a more intelligent way of combining images by taking the geometry/massing of one and applying the style/articulation of another, creating some truly striking results.

PromeAI and LookX have had tools in this category for a while. However, more recently Vizcom has released its own creative fusion tool that I can only describe as an absolute triumph of creativity. This will be covered in more detail in a future post, but one exciting application is when you want to apply the style of one project to a building with a totally different building type or size. To start, here’s an example of applying a unique art nouveau-inspired style to a drab looking apartment building using Vizcom:

For more examples of creative fusion, read AI Workflows for an Art Deco Revival and stay tuned for future posts.

Workflow Building – How to Combine Platforms

Brainstorming

Combining platforms requires the designer to determine the best AI image generator for each step of the design process. In general, it is best to start with either an existing image or one of the many high-creativity platforms to brainstorm.

Within your selected platform, it is often possible to supercharge your brainstorming and further develop an idea through remixing, image chain prompting, panning/zooming/outpainting, inpainting, prompt engineering, and various other tools which are are being introduced to the design landscape on a regular basis.

Conceptual and Schematic Design

After you have developed a concept you are satisfied with and want to set some of those design decisions in stone, the goal should be to move to a platform that offers more control and less creativity like one with constant composition (e.g., Playground AI), sketch rendering (e.g., Veras, Vizcom, or PromeAI), or creative fusion (e.g., Vizcom, Look.X, or PromeAI).

Once you start developing 3D models, you can continue to involve AI image tools through model screenshots (such as with Vizcom) or with plugins like Veras to explore design options across multiple views of your 3D model. This lets you refine the model as you go by having the AI focus on elevations, plans, or specific details to imagine what they might look like and offer rapid rendering capabilities. Tips and workflows for this type of model-centric design will be covered in a future post.

Next Steps

Finally we reach the end of what AI image generators can typically accomplish in 2023 (though that could easily change any day now). Future posts will further define these limitations, but this need not be the end of the road for AI-assisted design on your project. Other tools like LLMs, AI optimizers, AI code/performance analyzers, AI architecture rendering engines (see the link below to learn about my favorite free AI rendering tool) and other generative design tools can all play a role as the design progresses.

It’s an exciting new world for architectural design but with creativity and continuing education, the architects of today should have the tools to thrive in it.

Coming up…

In this article, we analyzed how the best AI image generators handle a modern art nouveau home using a predefined benchmark prompt. In the next post, we will use variations of this idea in an exploration of prompt engineering to experiment with the Midjourney algorithm, understand its underlying mechanics, and explain how to use those concepts to craft your own prompts.

Other posts you might like:

---

Subscribe to our Newsletter