Note: Aiello Studios does not earn revenue from links in this post.
Like many of you, we have been exploring using AI-generated images to create visual elements to use in our images. Our main questions have been, as fine art photographers, should we not consider using AI-generated images at all? Is there a way to take advantage of this new technology and not corrupt the responsibility for our own creativity?
At the outset, I want to clarify that we are not abandoning our regular image-making approach. Nor am I saying all photographers should use or not use AI-generated images. That's for each artist to decide. My goal here is simply to inform and provide a basis for thought and, perhaps, experimentation. As for us, we will occasionally use AI to create visual elements that we'll then modify, customize, comp, and blend into some of our in-camera images, primarily as backgrounds and textures. Any 100% generated images we make will strictly be for illustration purposes in blogs, social posts, workshops, or on our website and clearly labeled as an "AI-Generated Image, like the images in this post.
As we looked into this, we discovered we had some questions. How are these AI images made? Where does the source image material come from? Do we have the right to use generated images commercially? And if so, under what criteria? In this blog series, I will concentrate on and explore what I have learned thus far about making and using AI-generated images. So whether you're an aspiring photographer or a seasoned professional, join me as I explore AI-generated images and how you can use them in your photography. Today, we’ll begin with the main concepts and look at some of the players. In subsequent posts, we’ll delve into more detail about using AI image generators responsibly, including deeper dives into the ethical and copyright issues.
Technological Changes In Photography - A Bit Of Historical Perspective
From large wet plate negatives in the mid-18th century to modern digital cameras and sensors, the world of photography has constantly evolved. At each step in the advancement of technology, debates ensued. You may recall not all that long ago, some photographers boasted about not using filters, particularly when the then-new generation of effects filters came on the market. We debated the merits of digital image capture vs film. Many fine art photographers proudly claimed they never used (and never would use) Photoshop. Galleries marketed they would only represent photographers using silver or other "hand-made" prints (some still do, but they are becoming few and far between). Now, most of us, um, older photographers, no longer use our film-based cameras. We no longer have darkrooms. For many of us, advancements in digital photography technology greatly expanded our ability to be creative with color and, arguably, make better images than ever before. Technological progress, at least thus far, has allowed us to imagine, visualize, and create better than ever.
Now, the question is whether AI-generated imagery will take photography to new heights or be the end of human-created art. Will it accelerate the commoditization and devaluation of fine art photography? What is certain, from controlling cameras to helping photographers edit photos to creating stunning visuals, AI is redefining what we know, or thought we knew, about photography. Whether that's a good or bad thing is up to us.
Overall, notwithstanding some serious challenges, including ethical issues, I view developments in AI image generation positively because, like other technological advances of the past, the technology offers new tools that artists can use to express themselves in ways we could only imagine before. But yes, there are some serious ethical and copyright concerns as well. Not the least of which is the concept of fair use and attribution. I will not shy away from those concerns in this series of articles.
Getting Started with AI-Generated Images – A Guide for Photographers
For this series, I will explore AI-generated images as another tool in our virtual gadget bag to help us create images reflecting our vision. Understanding how AI image generators work can significantly enhance the results of implementing generated images into your photography. You can unlock a world of creative possibilities by identifying the right AI software or platform suited to your needs. Experimenting with different styles, techniques, and outputs allows you to create unique, innovative images that stand out. Learning how to integrate AI-generated images into your photography workflow, particularly in Photoshop, effectively can maximize their potential. With the power of AI image generators, you have the opportunity to transform your photography and, perhaps, elevate it to new heights.
AI Image Generators: An Overview
By utilizing artificial intelligence algorithms, image generators can produce stunning visuals, both realistic and not so much. Some of the top AI image generators in the market include DALL-E 2, DreamStudio, and Midjourney. These three tools offer a range of features and capabilities that allow users to create unique and striking images. They all offer free trials.
Experimentation and blending the generated images with your camera images can help you make the most of this innovative technology. Just remember, while incorporating AI-generated images into your photography workflow can save you time and effort while producing captivating results, it's essential to consider the potential drawbacks and ethical considerations when using AI-generated images. We’ll touch on these below and expand upon them in greater detail in an upcoming post in the series. A web / Google search quickly finds many other products and comparisons.
How AI Image Generators Work (non-technically speaking)
AI image generators learn from existing images. They don't create them out of thin air. The technology utilizes sophisticated algorithms to generate images based on vast datasets. The AI tools use these datasets to learn about the subject matter, like “oak trees,” “clouds,” “faces,” etc. The datasets do not actually store images. Instead, they are databases of image links (URLs) and captions that generative AI models use to “teach” their AI engines. The internet has made it easier for these images to be collected and, in most cases, used without permission. Some datasets get their data from licensed and non-licensed image sources. The non-licensed images are scraped from public web pages using web crawlers. This means they collect images (as links) from various sources, including social media, search engines, image-hosting websites, etc. This means all the image elements used to create your new image likely belong to the original creators. And that much of it is being used without their knowledge or permission. Hence, the potential copyright issues. This also means some of these “source” images may even be yours, mine, and those of our colleagues. I will cover ways to possibly exclude your work from the datasets in an upcoming article.
Using Prompts To Create And Customize AI-Produced Images
To generate an image, the user enters what is called a “prompt.” These are just simple phrases or sentences that describe the desired image for the text-to-image generator tool to analyze and render.
The more specific you are, the more likely the result will be something you can consider using. When you submit the prompt for processing, the tool will use what it has learned from the dataset to connect the meaning of the prompted words (e.g., “oak tree,” “daisies,” "road," “main street,” etc.) to visual elements in the dataset which it then combines and renders. The resulting output is usually in sets of 3 or 4 images for you to review, create variations of, and select from.
When you process a prompt, there is no guarantee that the results will be relevant, useful, or photo-like. To paraphrase Forrest Gump’s momma, “Generating AI images is like a box of chocolates. You never know what you’re gonna get.” So, generating several different images by modifying the prompt and resubmitting is not uncommon. Most of the tools refer to this as "making variations."
Editing
From what you may be accustomed to in photographic terms, editing is quite limited. As of this writing, the prompt is pretty much the extent of the inputs you can make to create and edit. In some tools, you can erase parts of the generated image, type in a revised prompt, particularly for what you want in the erased area, and generate a set of variations. This is called "inpainting." You can also use "outpainting" with prompts to extend the image's borders. As simplistic as it sounds, the process can be surprisingly helpful. Of course, you still have Photoshop for advanced editing and creating your final compositions.
Resolution
As far as resolution goes, the generated output is 1024 x 1024. Some have built-in resizing to approximately double the resolution when you download it. Still rather small. Therefore, a high-quality upscaling application like Topaz Photo AI or On1 Photo Raw is recommended for practical use. This may change as the technology advances.
Popular AI Image Generators
All three of the following web app AI image generators work well. At this point in time, the cost of all three is reasonably nominal. I suggest using the free trials to create images from the same or nearly identical prompts, then judge which works best for you. Prices are constantly changing, so the following fees may become obsolete so be sure to check their current pricing.
If you're looking for an easy-to-use art generator, DALL-E 2 by Open AI (also responsible for ChatGPT) is a great option. It is one of the best AI image generators available, but its user interface is overdue for an update. Resolution: 1024 x 1024.
Cost: 50 free credits, then $15 for 115 credits (or about $0.13 per image).
I like DreamStudio by Stability AI (which runs on Stable Diffusion) primarily for its interface. Although, it's a bit more challenging to get photo-realistic results. With this image generator, customization is a little easier. You can adjust various elements such as aspect ratio, color, style, and texture through the UI, whereas the other generators require commands to be added to the prompts. Additionally, from what I understand (so I may need to be corrected), Stable Diffusion only uses public web crawler image sources. So, there may be more copyright issues depending on how you use the results. Resolution: 1024 x 1024 that you can upsize to 2024 x 2024 within the app.
Cost: Free trial window, then $10 for 1,000 credits (good for 2-5000 images depending on various processing factors such as resolution)
Many say Midjourney's AI is the best, and I agree if you don't mind learning command parameters to customize the output you add to your prompts.
For example to output in a 3:2 aspect ratio need to add a command as follows. You can stack the command parameters one after the other for greater control.
"This is my prompt --ar 3:2"I don't like having to go through Discord, which requires a separate free membership and login. Not that I have anything against Discord; I just prefer SaaS apps to operate within their own environment and UI. Hopefully, that will change. Resolution 1024 x 1024. You can upsize to 2048×2048 (square) and 2720×1530 (widescreen).
Cost: After a free trial period, you can subscribe for $8.00 monthly and up.
We are currently testing the AI features of these tools and will report our findings in a separate post soon. They look promising thus far.
The Ethical Implications of AI Image Generators
I mentioned it above, but it bears repeating the ethical concerns surrounding AI-generated images should not be ignored. There are questions about authenticity, ownership, and the potential for misuse, including deep fakes, social bias, etc., especially in the age of social media and online artwork sharing. Not to mention originality. Therefore, as with any new technology, it is crucial to approach AI in photography cautiously. AI image generators should be used responsibly, ensuring that the images are not harmful, abusive, or lead to unintended consequences.
Copyright Considerations of AI in Photography
As with the ethical concerns when using AI-generated images in photography, it's also crucial to consider copyright implications and ownership. We have briefly discussed some of these above. Some AI image generators come with or are promising to have built-in copyright protections to mitigate legal issues. It may be there, but I have yet to see it. Photographers must know the terms and conditions surrounding using their AI-generated images to prevent potential legal disputes. Please carefully consider copyright implications when incorporating AI-generated images into your projects. I'll cover this vital topic more fully in an upcoming post.
Striking a balance between innovation and responsible use is crucial, and proper regulation may be necessary to ensure the ethical implications of AI image generators are addressed in the photography industry.
Conclusion
In conclusion, integrating AI into photography has opened up new possibilities and is transforming how we create images. But it is not for everyone. Integrating AI image generators like DALL-E 2, DreamStudio, and Midjourney has revolutionized the industry by offering innovative tools and techniques. However, it is essential to consider the ethical implications and copyright considerations associated with AI-generated images. As photographers, I believe we can embrace the AI revolution while being mindful of its potential pitfalls. By exploring the potential of AI in photography and understanding its place in our work, we can leverage this technology to elevate our creativity and produce stunning images that push the boundaries of our artistic expression.
Sounds like fun to me! Watch this space for more articles on this topic. Let me know what you think.
Of course, I invite you to subscribe to our newsletter from our contact form, or our Facebook page to be notified when I post additional articles and updates.