Generative AI: Fake it ‘til you make it!
By Gabrielle Stannus
Are you the only one in your circle who is not yet using AI? Are you wondering how it may be incorporated into your horticultural or landscape design business? Generative AI may be able to help you create text or images to use in your business operations (administration and marketing) or design (content and visualisations). Read on to find out what is happening in this space, and how you can get a start on using some of these tools to save you time.
I bet that many of you did not get into horticulture and/or landscape design to spend your time in front of a computer. However, the reality is that many of us now spend a good part of our day tapping away at a keyboard. So, how can we reduce the time spent in front of a screen by making those keystrokes just that little bit more effective? Whether you want to write a blog or social media post to market your plants or generate images to inspire your creative processes when designing a new landscape, generative artificial intelligence (AI) can make these and many other tasks easier.
The Oxford English Dictionary defines artificial intelligence as “the capacity of computers or other machines to exhibit or simulate intelligent behaviour; the field of study concerned with this”1. Generative artificial intelligence (AI) is a type of AI that generates images, text, videos, and other media in response to prompts inputted into those computers or other machines.
You are probably already using generative AI in your daily work without even realising it unless you have been hiding under a (nicely landscaped) rock. If you use Microsoft (MS) Office’s Word program to prepare documents, then you have likely used the Editor’s spelling and grammar check to communicate more clearly and concisely, and perhaps even its rewrite function as a result? Even as I type this article in MS Word, I am continually prompted by this program to auto-fill words to complete the phrase or sentence I am writing.
You have also likely come across what are known as ‘predictive’ chatbots providing frontline customer service. These virtual or digital assistants include ‘Siri’, found on Apple’s iPhone Operating System (iOS), Amazon’s ‘Alexa’ and Google’s perhaps less originally named ‘Assistant’. My favourite chatbot is ‘Alex’ on the Australian Taxation Office’s (ATO) website. I obviously lead a very exciting life! If you run a retail nursery or sell horticultural goods online, then you may already have your very own website chatbot assisting customers with their inquiries, thereby saving you time.
These chatbots all rely on deep learning algorithms, a type of machine learning. Machine learning is a subset of artificial intelligence and refers to the study of computer systems that learn and adapt automatically from experience without explicitly being programmed2. Machine learning algorithms usually require human correction when they get something ‘wrong’. If you are a user of a music or video-streaming service (e.g., Spotify), then you will know that those services learn your preferences to offer you listening or viewing suggestions. Mostly, these suggestions will be to your liking, but sometimes not, so you will tell that service that you do not like that suggestion, so it learns not to offer similar suggestions to you in the future. Deep learning is a machine learning technique that layers algorithms (sets of rules) and computing units (neurons) into artificial neural networks that mimic the human brain. Deep learning algorithms can improve their outcomes through repetition, without human intervention2. These algorithms are being put to good use in a new ‘breed’ of chatbot.
New kids on the block
Over the last year, the world has become more than a little excited by the release of OpenAI’s ChatGPT, followed by Microsoft Bing Chat and Google Bard chatbots. These computer programs are trained to follow instructions and provide a detailed response to a prompt submitted to them by their users – you and me! A prompt is essentially a question that you ask of the program, i.e., you ‘prompt’ it to do something. Prompts are also known more formally as natural language descriptions. ChatGPT and Google Bard will answer your questions responding only in text, whereas you can prompt Bing Chat to answer your question using text and it can now also generate an image for you using DALL-E 3, but more on that later.
Generate text
As well as answering your questions in an informative way, these chatbots can provide summaries of factual topics, create stories, and translate languages. My French partner finds it very helpful to use a chatbot to help him write professional emails to ensure that he uses an appropriate tone of voice and that his spelling and grammar is correct.
Generative AI is very helpful when you would like an overview of a topic that is new to you. For me that topic is generative AI! So, I put ChatGPT to the test to see how it could help me summarise the use of AI in landscape design. I ran this question through ChatGPT using this rather rudimentary prompt exactly as it is written here: “How can AI be used in landscape design?”.
I have shared the response I received from ChatGPT verbatim in the “How can AI be used in landscape design” pullout box in this article. Spelling and grammar pedants (and the HJ Editor) please note that American English is the default in ChatGPT!
How can AI be used in landscape design?
Artificial intelligence (AI) can be used in landscape design to enhance the efficiency and creativity of the design process. Here are several ways AI can be applied in landscape design:
Site Analysis and Data Collection:
AI can help gather and analyze data about the site, such as soil quality, topography, climate, and existing vegetation. This data can inform the design process, ensuring that the landscape design is well-suited to the environment.
Generative Design:
AI-powered generative design tools can create multiple design alternatives based on specific parameters and constraints. Landscape designers can explore a variety of design options quickly and efficiently.
Simulation and Visualization:
AI can be used to simulate how a landscape design will evolve over time. This can help designers and clients visualize how plants will grow, how the design will change with the seasons, and how it will interact with the surrounding environment.
Plant Selection and Recommendations:
AI can assist in selecting appropriate plants and materials for a landscape design. It can consider factors like local climate, soil conditions, and the client’s preferences to suggest suitable plants and materials.
Irrigation and Water Management:
AI can optimize irrigation systems by monitoring weather conditions and soil moisture levels in real-time. It can adjust irrigation schedules and water distribution to ensure the efficient use of water resources.
Maintenance and Care Recommendations:
AI can provide recommendations for ongoing maintenance and care of the landscape, including when to prune, fertilize, or treat for pests and diseases. It can also offer guidance on seasonal tasks.
Project Cost Estimation:
AI can help estimate the costs associated with a landscape design project by analyzing materials, labor, and other expenses. This can aid in budget planning and cost management.
Accessibility and Universal Design:
AI can assist in designing landscapes that are accessible to all individuals, including those with disabilities. It can recommend features like ramps, paths, and seating arrangements that meet universal design standards.
Environmental Sustainability:
AI can help in designing landscapes that are environmentally sustainable. It can recommend eco-friendly features like rain gardens, green roofs, and native plantings that promote biodiversity and reduce environmental impact.
Feedback and Learning:
AI can continuously learn and adapt from previous landscape design projects, improving its recommendations and designs over time based on the success and feedback from previous projects.
Collaboration and Communication:
AI tools can facilitate collaboration between landscape designers, architects, and clients by providing a shared platform for design collaboration and real-time updates.
AI in landscape design is not meant to replace human designers but rather to augment their capabilities and streamline the design process. It can save time, reduce errors, and provide valuable insights that can lead to more functional and aesthetically pleasing landscape designs.
(NB. This pullout box content was created with Open.AI ChatGPT using the prompt: “How can ai be used in landscape design?”, on 6 November 2023. The response has been printed here verbatim.)
Generate 2D images
In addition to creating text, some of these chatbots and other generative AI platforms can also generate images which may inspire your landscape design thinking and aid in concept design. You can use text, image prompts, or image to image, to create new 2D images using AI models hosted and/or developed by Bing, Midjourney, DreamStudio and Adobe. Just like the chatbots discussed earlier, you need to prompt these AI models to generate the content that you want to see.
My first attempt to generate images using AI was via Midjourney. This generative AI model is only accessible through Discord’s Midjourney server. Discord is an instant messaging and voice over internet protocol (VoIP) social platform whose users can communicate with voice calls, video calls, text messaging, media, and files in private chats or as part of communities called “servers”. Free use of the Midjourney AI has been restricted in recent times as the Midjourney server has been overwhelmed by demand for its services.
More recently, I have had a little more success generating 2D images from text using Leonardo.AI. This platform is currently my building designer partner’s personal favourite as it has several free to use AI models to choose from, in addition to pay for use options, which generate high quality images (depending on the prompt you put in of course!). The AI models available here include free versions of Stable Diffusion, an open-source program created by Stability AI and its partners, which can be used for architectural and other renders.
However, if generating images from text is all rather new to you, I suggest you start with the Bing Image Creator. Powered by DALL·E 3, this is perhaps the most accessible of the AI models currently available as it is free to use if you have a Microsoft account. Just open the Microsoft Edge internet browser and click on the blue-green icon in the top right-hand corner to open the Bing Chat Copilot and enter your first prompt. Note that you may have to download and install the Edge browser first if it is not already set up on your computer. That should only take a few minutes. For best results using DALL·E 3 to generate images, I would recommend that you go straight to the source, i.e., DALL·E 3’s creator OpenAI (yes, the same ‘guys’ responsible for ChatGPT), but you will have to pay for the privilege!
Generate 3D images
And now I get to the very exciting part! Photorealistic text to 3D image generation is here, well at least prototype versions are now available. Luma AI recently released Genie, a 3D foundational model available as a research preview on Discord. If you are an early adopter of technology, I highly encourage you to check it out. On the diffusion of innovations bell curve, I consider myself part of the middle majority when it comes to taking up new technology. So, I think I will wait until my partner plays with Genie and reports back to me more fully on its merits with regards to 3D visualisations.
The early adopters in this audience may know that SketchUp now has an AI-powered visualisation tool called Veras. Designers can use text prompts to ask Veras to create photorealistic renders of their designs. Similarly, Arko.ai also promises to transform 2D sketches and models into realistic renders. The makers of Arko.ai claim the use of this model is compatible with SketchUp, Revit and Rhinoceros.
Integrate AI-generated images into existing photos
My partner speaks very highly of Adobe Firefly – – Generative AI for Photoshop and its ability to integrate AI-generated images into existing photos. He describes it as “Photoshop for people who have no clue how to use Photoshop”, meaning that it involves little or no skill to use. You can prompt the different tools within this pay– for– use program to generate images from a detailed text description (text to image), use a brush to remove objects of paint in new ones (generative fill), or apply styles or textures to words and phrases (text effects).
Pay another professional to use AI to design on your behalf
Some professionals are now promoting integrated ‘concept to reality’ design services online. Using generative AI and other programs, these people claim to be able to create detailed landscape designs in minutes. Generally, this type of platform will ask you to upload a photo of the site which needs landscaping, then select your preferred garden style (e.g., drought resistant garden), before choosing the type of outdoor furniture and other structures that you would like to see incorporated into that garden. Want to know more? Check out the following websites: HomeDesignsAI (image to image generation), DreamzAR (image to image), Fotor (text to image). You will pay for these services of course!
………………………………………………………………………………………………..
How should I acknowledge that I have used AI to generate an image?
Acknowledging the use of AI to generate an image is important to provide transparency and give credit to the technology and tools used in the creative process. Here’s how you can acknowledge it:
- In the image caption or description: When sharing the image, you can include a brief mention in the caption or description, such as “This image was generated with the assistance of AI technology” or “AI-assisted artwork.”
- In the image metadata: If you’re sharing the image on a platform that allows you to include metadata, you can add information about the AI tool or software used in the generation process.
- In the creative work itself: If you’re creating a digital or physical portfolio, you can include a section in your portfolio that discusses the use of AI in your creative process and acknowledges the AI tools or models used.
- In accompanying documentation: If you’re submitting the image for a competition, publication, or exhibition, make sure to include information about the AI technology used in your submission materials, artist statements, or supporting documentation.
The specific wording may vary depending on the context and your personal preference. The key is to be clear and transparent about the role of AI in generating the image while giving credit to the technology or tools that contributed to the creative process.
(NB. This pullout box content was created with Open.AI ChatGPT using the prompt: “How should I acknowledge that I have used AI to generate an image?”, on 6 November 2023. The response has been printed here verbatim.)
The final prompt!
If you are worried that AI might take your job, don’t be! Some people following the development of generative AI more closely than I am, claim that it is very likely that there will be some skill displacement; however, they are sceptical that job displacement will occur, except in those niche, narrow roles3 . My partner’s very detailed prompt seeking help to generate a landscape image in ChatGPT saw him receive the following response from that AI model: “Please consult with a local landscape designer or horticulturist for precise plant placement and care instructions based on your region’s climate and soil conditions.” I could not have replied any better myself!
There is still a place for us humans in this brave new world;, landscape designers and horticulturists alike. Generative AI is evolving rapidly. Many of the platforms I have written about in this article provide free AI models that you can ‘play’ with. So, get in there and have a go. Explore what they have to offer and see whether they may help you to develop and grow your business more effectively. To help you along the way, find your AI tribe, whether that be your partner, child (yes, teenagers are an excellent source of information), colleagues, employees, or an online community. Most importantly, remember when asking the chatbots for help, that like humans,; their response is only as good as the question put to it. So be nice to, and patient with, them!
TIPS TO HELP YOU MAKE THE MOST OF GENERATIVE AI
Learn how to ask the right questions
The quality of the output generated by these AI models is only as good as the input. Think of a chatbot like a librarian. That person can only help you find the information you really need if you ask them the right questions. The way that each AI model will dip into their ‘library’s catalogue’ depends upon how you frame your prompt to them. My partner has sought help in this area by writing a prompt telling ChatGPT that “I want you to act like a prompt engineer”. There is a little more text than that in my partner’s prompt. However, rather than bore you with that detail, here are some tips for writing more effective prompts for use on chatbots as recommended by ChatGPT and summarised by me:
When writing a prompt, be clear and specific about what you want the chatbot to do. Avoid vague and ambiguous language. Provide context and use keywords. Experiment with phrasing if you are not getting the response you want, and/or consider breaking down a complex prompt request into multiple steps.
For text generation, ask open-ended questions, specify the tone and style you would like the response to be written in, and provide examples of the type of response you are looking for.
For image generation, include descriptive details, use adjectives and adverbs, and reference existing images when writing your prompt. Experiment with style and genre to make sure you get an image you are happy with.
Still stuck coming up with a prompt after following these tips from ChatGPT? Then try a prompt generator. Some AI models have in-built prompt generators that you can use. Otherwise search online for some help.
Alternatively, consider using a ‘negative prompt’. Tell the chatbot what you do not want, i.e., “I want this NOT that”. This function is built into many generative AI platforms, e.g., Leonardo.AI. Click the appropriate button to activate that prompt.
Avoid ‘over-sharing’
Like other online platforms, be very mindful of what you share as it may enter the public domain. How will your personal data be stored and who will have access to it? Avoid entering personal or confidential information into your chatbot ‘conversations’. Google Bard explicitly warns its user not to enter any data that you would not want a human reviewer to see or Google to use to improve their products, services, and machine-learning technologies.
Choose your own adventure!
On most of these platforms, there is the option to train your own AI models to generate images in a style of your choice by uploading images that inspire you. Perhaps a famous landscape architect like Roberto Burle Marx? Or a particular plant palette that you are fond of, e.g., Australian landscape? Think big!
Add the human touch
Content created using a generative AI model may lack the human touch, with some text being somewhat dry and at times repetitive. There is, and will remain, a place for human writers and content creators, with all our idiosyncrasies, quirks, and foibles! So, use these chatbots sparingly and make sure to bring your personality to your writing or design. Think of generative AI as a research assistant, but not as your replacement.
Select a platform with ethically sourced or authorised reuse content
Generative AI is a relatively new area of commercialisation. Legislation regarding the appropriate reuse of content in AI models differs around the globe. The recent strike by the United States’ Screen Actors Guild and the American Federation of Television and Radio Artists (SAG-AFTRA) was concerned in part with the use of generative AI to simulate a performer’s voice, facial expressions, and movements to create entirely new content without the performer’s explicit consent.
Unlike many AI tools, Adobe’s Firefly was trained using only licensed images and public domain content. This makes it a brand-safe option for corporations and commercial design work3. OpenAI has recently said on its website that DALL·E 3 is designed to decline requests that ask for an image in the style of a living artist. Creators can now also opt their images out from training of our future image generation models. Generative AI by Getty Images also claims that it only uses licensed images to train its AI model, and that its content creators are compensated for the use of their work accordingly.
Don’t rely on only one source!
Where do these chatbots get their information from to answer your questions? After prompting it, Bing Chat’s Copilot ‘told’ me that it gets its information from various sources on the web, such as news articles, websites, blogs, social media, etc., and that is uses Bing to search the web for relevant and reliable information. It also claims to use its own knowledge and creativity to generate content that is not available on the web. ChatGPT ‘told’ me that it was trained on a mixture of licensed data, data created by human trainers, and publicly available data including books, websites, and other texts. So, keep in mind that some of the information you may receive from a prompt answered by a chatbot may be biased. For example, if you ask a chatbot to recommend a new computer for your business, is it really giving you the best advice, or like some search engines, are some results prioritised over others? Will some content be sponsored in the future? Keep an open mind and cast a critical eye over developments in this space.
Acknowledge your sources
AI generated content does not currently come with its sources acknowledged as you may have noted in the How can AI be used in landscape design? pullout box in this article. Google Bard acknowledges this, encouraging its users to ‘fact check’ using the Google search engine. Academic integrity is a very important part of my other work teaching geography at the University of Tasmania. I am constantly reminding students that they must acknowledge the work of others by citing the sources they have used to provide evidence to support their arguments. It would be remiss of me not to remind you to verify the claims made by the chatbots you converse with, and then to provide appropriate credit to those sources. A good chatbot comes with a warning: they ‘themselves’ admit they can make mistakes4!
…………………………………………………………………………………………….
Gabrielle Stannus
Inwardout Studio
M: 0400 431 277
E: gabrielle@inwardoutstudio.com
References
- Oxford University Press 2023, artificial intelligence, Oxford English Dictionary, viewed 14 November 2023, https://www.oed.com/search/dictionary/?scope=Entries&q=artificial+intelligence
- Coursera 2023, Deep Learning vs. Machine Learning: A Beginner’s Guide, Coursera, 30 October 2023, viewed 14 November 2023, https://www.coursera.org/articles/ai-vs-deep-learning-vs-machine-learning-beginners-guide
- Price, A 2023, Text-to-3D-generation is here, This Week in 3D e-newsletter from Andrew Price, 3 November 2023, viewed 15 November 2023
- OpenAI 2022, Introducing ChatGPT, OpenAI, viewed 14 November 2023, https://openai.com/blog/chatgpt
NB. All images generated by Ludovic Vilbert, Inwardout Studio, using AI models found freely on Leonardo.AI. Only portions of the prompts used to generate these images have been shared here in the interests of brevity.
Main photo: A photo-realistic landscape design featuring a front garden. The prompt requested that the image be “infused with the style of Claude Monet’s works, evoking the dreamy atmosphere of his Giverny garden” (AI model used = Absolute Reality v16)