Breakdown: Why the Same Query in formspace.design Yields Different Results

When I first started working with generative models, one thing struck me as an artist and someone with 25 years of experience in landscape design: giving the same query, I received different, yet always interesting, results. This isn't an error, but a fundamental feature of AI's operation, which I actively use in formspace.design. Today, I'll explain why this happens and how to get the most out of this "unpredictability."

Why AI is Unpredictable: The Internal Mechanics

Imagine a neural network as an artist who interprets your request slightly differently each time. This happens due to several factors. Firstly, many generative models use elements of randomness (so-called "seeds" or "seeds of randomness") to initialize the generation process. This helps avoid monotony and creates unique images with each run. Secondly, even with the same "seed," minor changes in the computational environment or even in the data processing sequence can lead to subtle but noticeable differences in the final result. For me, as someone who once painted, this is very similar to how an artist might paint the same landscape several times, and each time it will have its own unique character.

Thirdly, AI doesn't operate with "concepts" in the human sense. It works with patterns and probabilities, and each time it generates an image, it does so by choosing from billions of possible options based on its training. It's like asking a thousand different people to draw a "beautiful garden" – each will present something unique, but all will conform to the general idea of a "beautiful garden." It is this "creative" interpretation that makes the process so fascinating.

My Work with Variations: From Painting to Landscape

My journey in landscape design began long before I started coding with AI at 61. I always sought new approaches, whether in the "Alkar-Invest" business in the 90s or in my company "Dom i Sad" (Home and Garden). Even as a student, I painted and even exhibited at VDNKh. This "artist's eye" taught me to appreciate diversity and seek new ideas. When I was first learning AutoCAD, then Revit and 3ds Max for design, creating a single sketch took up to 3-4 hours, and making revisions was laborious. With the advent of AI, this process accelerated many times over. Now I can get dozens of options in a matter of minutes.

Instead of striving for absolute repetition, I use this variability as a powerful brainstorming tool. For example, if I want to create a "conifer corner" on a 15-20 m² plot, I can give the AI the same query, but each time receive different compositions with various types of conifers, stones, and decorative elements. This allows me to quickly explore many ideas that would otherwise take days or even weeks of manual work. For me, this has become a natural continuation of the creative process, only now with incredible speed and volume of idea generation. I've already discussed how I teach the neural network to understand my queries for a conifer corner in the article How I explain to the neural network what a garden's conifer corner is.

How to Formulate a Query to Get What You Need?

Formulating a query (prompt) is an art that determines how useful the variations will be. I've found that to get the desired result, it's important to balance specificity with some freedom for the AI. If your query is too rigid and detailed, you might get very similar results, but you'll miss potentially interesting and unexpected solutions. On the other hand, too general a query can lead to irrelevant results.

My strategy is this: start with a general description, for example, "Mediterranean-style garden with a lounge area." Then, if I want to see more variations, I can add small but important details: "with a stone gazebo," "with gravel paths 1.2 meters wide," "with an emphasis on lavender and rosemary." Or, conversely, I can vary the main query by changing just one or two words: "Mediterranean garden, evening lighting" or "Mediterranean garden, morning fog." This allows the AI to reinterpret the task, but within the given style. I wrote more about prompt composition in the article Step-by-step: how I create a prompt for garden visualization from scratch.

Common Mistakes When Working with AI Generation

One of the most common mistakes I see beginners make is expecting absolutely identical results from the same prompt. This leads to disappointment. AI is not a copy machine. Another mistake is overly long and convoluted prompts that contain contradictory instructions. AI can get confused and produce something meaningless. I always advise breaking down complex ideas into several simple queries.

It's also important to remember that AI doesn't read your mind. If you want a specific type of plant or a particular material, you need to state it explicitly. For example, instead of "beautiful flowers," write "'Annabelle' hydrangea bushes." And finally, don't be afraid to experiment. Sometimes the best ideas come from the most unexpected options that AI generated when you least expected it.

When Variability is an Advantage, Not a Problem

For me, variability is not a problem, but a huge advantage. It is especially valuable during the conceptual design phase when you need to quickly sketch out several different ideas for discussion with a client or for your own inspiration. Let's say I have a 10-are plot, and I need to propose three completely different zoning concepts – for example, one with a large pond, another with a landscaped lawn and a children's area, and a third with a vegetable garden and an orchard. With AI, I can get rough versions of all three concepts in a short time, which would have taken days before.

Variability also helps when you're looking for an unusual solution. For example, I can experiment with different lighting styles or paving options for paths 1-1.5 meters wide. AI can suggest combinations that I would never have thought of myself. This is especially useful when you need to go beyond familiar patterns and offer something truly unique to the client. For instance, for a barbecue area of 25-30 m², you can get options with a fire pit, a large oven, an open kitchen – and each time AI will offer its unique arrangement and design of elements.

How to Plan in formspace.design

At formspace.design, we built a tool that I myself lacked in landscape design practice. I understood the pain of a director who wants to quickly get different ideas and not spend days on a single sketch. That's why the AI generation functionality (Module 2.1 AI Concept) is the heart of our planner.

Imagine you want to create a lounge area on your plot. You can start by uploading your site plan in Module 1.0 "Site Plan" and marking the boundaries of the future area. Then, go to Module 2.1 "AI Concept." In the prompt field, you can enter something like: "Modern style lounge area, with a small fountain, comfortable chairs, and low decorative shrubs." Click "Generate." You will see several options, each of which will be unique. If you like an element in one option, but the overall composition is better in another, you can repeat the query, slightly changing the wording, or use Module 3.1 "Design from Photo" to refine the chosen concept by overlaying it onto a real photo of your plot. This allows you to quickly and iteratively create and improve designs without being a professional designer or programmer. This is how I, not being a programmer by education, build formspace.design, using AI to create functionality that I lacked in my practice.

FAQ

Can I get an absolutely identical result in formspace.design?

In most cases, this is not possible. The AI models used in formspace.design, like other generative neural networks, include elements of randomness in the image creation process. This is done to offer unique ideas each time and prevent monotony, stimulating creative exploration.

What if I don't like any of the generated results?

If the initial results don't meet your expectations, try to refine or rephrase your query. Add more details about the style, plants, materials, or elements you want to see. Sometimes, changing just one or two words is enough to get a completely new set of ideas.

Does the length of the query affect the degree of variability?

Yes, the length and detail of the query can affect variability. A shorter and more general query usually gives the AI more freedom for interpretation, which can lead to more diverse results. Very long and specific prompts, on the other hand, can narrow the range of options, making them more similar to each other.

How many times can I generate a design using the same query?

You can generate a design using the same query as many times as you wish. formspace.design allows you to freely explore various options. Each time, you will receive new ideas that will help you form the ideal landscape project.

Is this an AI feature or a technical error in formspace.design?

This is not an error, but an inherent feature of how generative AI models work. At formspace.design, we value this variability as it expands creative possibilities and helps our users find unique and unexpected solutions for their plots.

Can I combine the best elements from different generated options?

Absolutely! This is one of the most effective ways to work with AI. You can take preferred details from one option (e.g., the shape of a pond) and a zoning idea from another, then use them as a basis for further refinement in Module 3.1 "Design from Photo," overlaying elements onto real photos of your plot.

How do I choose the best option from the many proposed by AI?

When choosing the best option, rely on your feelings and practical applicability. Consider how well the proposed design meets your needs, the style of your home, and the characteristics of your plot. Sometimes it's helpful to show several options to close ones to get a fresh perspective.

Unlocking the Creative Potential of AI

My experience shows that AI doesn't replace human creativity; it enhances it. The variability offered by formspace.design is not a bug, but a feature that opens new horizons in landscape design. Use it to explore, experiment, and find unique solutions for your plot. I believe that with this tool, every plot owner, whether it's 6 ares or 20+, can create the garden of their dreams, even if they've never held a designer's pencil. Just start experimenting.