← Go Back

Stable Diffusion is a tool you need to use

Stable Diffusion has taken the world by storm. In 2022, OpenAI released its second iteration of DALL-E, the text-to-image generator. While it was very impressive, it was closed source and behind a paywall. That meant that the use-cases and media coverage of it was still limited. In August 2022 though, everything changed when Stability AI released its own diffusion based image generator: Stable Diffusion.

Stabolity AI Logo

Unlike its predecessors, Stable Diffusion was Open Source, Free and public under a pretty permissive license, allowing anyone with the necessary computing power to generate a vast number of images for free. It also meant that users could modify it to their liking and teach it new concepts (another article on this is incoming).

And thus, the AI images revolution began. A plethora of tools from face swapping to image generation on demand flooded the market. We are still in the wild west phase, and a lot of these tools are unrefined, but there is already a clear learning we can take from this: you need to get confortable with these tools if you want to continue being productive. Whatever your job, it can be done faster and more easily with tools like Stable Diffusion or GPT-3. It will soon be a necessary skill to be comfortable with AI tools and know how to incorporate into your workflows.

Stable Diffusion for artists

Artists first widely protested against the tool, criticizing how art was used to train models without their consent. I am not going to enter this debate here, but just know that there are legal battles underway. Now, even if artists would win the case against Stability AI, the fact that it's open source mean anyone can still use their work to create a new model and there is little to nothing they can do about it. Even a law forbidding it would be virtually unenforceable, as it's impossible to know how a model was trained if their creator does not want to disclose it.

To me, that means that those tool will continue to exist, grow and thus are now part of the life of artists. Some accepted that very quickly and instead used the tool as a way to improve their craft. They used it to iterate over drafts and ideate: for example, you can feed Stable Diffusion an image and generate plenty of alternatives. By doing so you can quickly check and evaluate how some changes would affect your creation and then pick your favorite to work upon.

Example of variations you can generate

You don't really have an idea for a composition and just a vague idea of a concept ? No problem: generate several images with Stable Diffusion and get inspiration to create your own piece. You can also use it to generate reference images of buildings, animal, scenery, and others.

Best of all, SD is open source and can be modified to your liking. It can learn your unique art style and generate some crude references for you to build upon. Instead of creating a few small drafts, you can instantly generate a few hundreds and use them to create your final piece, cutting down on the ideation phase.

Now imagine two artists. One is just using his previous workflow, while the other leverage AI to accelerate its ideation phase and test different versions of his work. The first will take a long time to go back to a client, while the second one can send numerous drafts to get feedback on what the client prefers. Imagine exchanging with a tattoo artist and instead of going back and forth for days modifying, the artist can immediately send you several rough drafts to pick from before working on any of them.

The gain in productivity, customer satisfaction, and reduction of scrapped iterations makes it a very valuable tool. All this while its cost is nearly none: most digital artists have a computer with a beefy GPU (3070 or higher) and could generate images directly on their own machine. I believe that those able to leverage these tools will quickly see increases in productivity, revenue and ease-of-life, while those who do not will fall behind.

Stable Diffusion for bloggers or developers

Artists are not the only one that will have their work impacted and that can leverage Stable Diffusion: developers, bloggers, social media manager are also faced with an opportunity.

Finding images to illustrate your mockups or articles can be a challenge. You have free database of quality images like Unsplash, but they lack originality or might not totally fit your content. How about, like in this article, being able to generate several versions of the illustration you are looking for, free of any complex license or copyright ?

For developers, you could even imagine a library that do that for you. Imagine building a website for a client and needing placeholders for resources. Will you spend the time necessary to find, collect, optimize and then resize many images ? How about instead just writing <AiImage desc="a photo of an astronaut in ordbit around earth" size="..." /> and getting an image generated directly with the appropriate accessibility decorations, optimized for the size of your website, and free to use. You don't like what was generated ? No issue, just click the small icon on the image and get a new one.

The ability to save time on repetitive, low-value tasks is what will set you apart from other developers. How do you feel clients will react when you are twice faster than other engineers ? They will be delighted, unaware that you are not a coding genius that spit out code twice faster than other devs, you just leveraged open source tools to cut down on the amount of time spent on tasks that did not leverage your core skills.

Whoever you are, it's coming for you

The two examples above are just the ones I am the most comfortable talking about as they are close to my daily job, but they are not the only ones.

So many jobs will require illustrations or inspiration. A stylist could generate variations of outfits and iterate on their creations. An architect could generate some concept illustration for its proposal. A 3D artists could generate on demand unique texture for its creations... The possibilities are endless.

How about generating complete new products

Being an early adopted and learning to leverage these tools will thus clearly give you an advantage. I believe that in most white collar jobs, Stable Diffusion or an equivalent will become part of the daily job and essential to stay competitive. So why not start to get used to it now ?