Have you heard of AI? It’s kind of a big deal right now, from writing essays to creating "art" automatically and in a matter of seconds. Is this the end of humankind, or a revolutionary tool for efficiency? We'll look at some of the major players in AI image generation and see how they might help inspire landscape designs. You’ll learn how to access these tools and pick up some tips for generating images to help with conceptual design.
Note: The following catalog of content covered in this webinar is time stamped to allow you to follow along or skip to sections of the video that are relevant to your questions. You can also search for content on this page using the FIND command in your browser (CTRL + F in Windows, Command + F in Mac OS.)
- AI Image Generator Overview
- What Exactly is an AI Image Generator?
- How Does It Work?
- Various Options Available for Generating Images Using AI
- Midjourney Tools for Generating Images
- DALL•E Tools for Generating and Editing Images
- Food for Thought
0:00 – 6:09: Intro/TOC
6:10 – 22:35: AI Image Generator Overview
What Exactly is an AI Image Generator? (6:10)
What is AI art or an AI image?
- AI Images are created using machine learning algorithms, also known as Artificial I
- These are used to produce unique images.
- Text-to-image AI image generators are one type of AI art.
- With this technology, users can simply type in keywords and AI will generate an image based on the provided description.
How Does It Work? (7:00)
- Training an AI model to create unique images requires vast amounts of data.
- Millions or even billions of images are entered into the model, along with accompanying descriptions.
- The AI uses this data to make associations between the description and the images.
- When given a description, the AI produces a unique interpretation by referencing the associated images.
Various Options Available for Generating Images Using AI (7:37)
We will be looking at the current top 3 AI image generating models:
- DALL•E 2
- Stable Diffusion
Midjourney (beta) (8:02)
- Open beta started July 12, 2022.
- Rendering done through cloud services, so you don't need a beefy graphics processor.
- No standalone user interface, yet. Instead, users can access through Discord (a free communications app, available at discord.com).
- Best for high-quality image generation without being too precise with descriptions.
- Version 1 of Midjourney was released in July 2022.
- Since then, a new version has been released every few months.
- With each update, the results generated by Midjourney have become increasingly detailed and accurate.
What can you do with Midjourney?
Text-to-image generation by typing /imagine and then your image description.
- Each time you generate images, you get four variations to choose from.
- To narrow down the look you want, create four more variations based on the original prompt or one of the original four images.
- Once you've found the right image, upscale it to get a higher quality version of the image.
- Upload reference images to influence results.
- Blend multiple images together for a single image based on how the AI interpreted your uploaded images.
- Have it describe an image that you upload.
Accessing Midjourney (9:41)
- DALL•E originally announced/ released in January 2021 from OpenAI.
- A new and improved version, DALL•E 2, was released in September 2022.
- Outsources rendering to GPU farm so you don't need a beefy graphics processor.
- Various applications available for utilizing the DALL•E engine.
What can you do with DALL•E 2?
- Text-to-image generation (Describe an image that you want and let DALL•E create it.)
- Upload and edit images
- Inpainting ("Inpainting is the ability of DALL•E to fill in missing parts of an image based on contextual information." – Laura Carnevali with medium.com)
- Outpainting ("Outpainting is the ability of DALL•E 2 to generate images that go beyond the boundaries of the prompt, extending them naturally and seamlessly." – Laura Carnevali with medium.com)
Stable Diffusion (19:00)
- Developed by Stability AI and released in 2022.
- The model is available through various applications, but lacks a first-party, user-friendly interface.
- Versatile in application.
- Free to use.
How can you access Stable Diffusion?
Similar to Dall-e, it is open source, so there are several ways to use it:
- Stable Diffusion v1-5 Demo (co/spaces/stabilityai/stable-diffusion)
- A web-based interface for trying Stable Diffusion. Easy to use
- Popular advanced graphical user interface (GUI) for Stable Diffusion.
- Somewhat technical to set up, but there are several online guides available for installation and usage.
- Can use your computer's onboard graphics processor or outsource graphical processing to Google Colab.
- Other applications such as DiffusionBee (Mac)
- Paul used DiffusionBee to check out what Stable Diffusion can do. His results weren't great; they seemed limited, and the user interface wasn't very friendly – similar results to DALL•E but lower in quality.
What can you do with Stable Diffusion?
- Text-to-image generation
- Image based on your images
Note: Based on Paul's experience with DiffusionBee, it seems that interfaces like AUTOMATIC1111 offer more control over the results by adding parameters. He plans to explore this further, but this seemed outside the boundaries of this introduction-style webinar.
22:36 – 51:01: Demonstration
Midjourney Tools for Generating Images (22:50)
DALL•E Tools for Generating and Editing Images (43:16)
51:02 – end: Food for Thought
- At present, AI-generated images cannot be copyrighted.
- This has resulted in many gray areas surrounding the legal ownership and use of AI art.
- Despite the legal uncertainties, there are still valuable takeaways from these images.
- Color schemes, forms, and new ideas can all be derived from AI-generated art.
- Focus on the elements that suit your environment, and incorporate them into a practical design.
- AI-generated images have improved significantly in recent years.
- It is expected that these advancements will continue, resulting in even better image generation.
- User interfaces will become more intuitive and user friendly.
- The future of AI art is uncertain due to potential regulations brought on by ethics disputes and lawsuits.
- Is this the death of art and creativity? Are designers about to lose their jobs to AI?
- AI is not yet advanced enough to replace human creativity and ingenuity.
- Artists and creative minds are still needed to guide and direct the AI.
- Human understanding and problem-solving skills are still necessary for successful design.
- It is unlikely that AI will completely replace human creativity in the foreseeable future.