Should artists render AI images on their own hardware?

Image created by PJ Brunet

Another big day in AI.

– Adobe has announced Firefly, an AI tool for creatives
– Google opened up BARD LLM
– NVIDIA launched cloud tools for Generative AI
– Microsoft announced Bing Image Creator

More than just image generation (like Midjourney or Stable Diffusion), Firefly allows you to create and EDIT images, create text to 3D, text to Vectors, upscale and extend images, etc.

Responding to this post on LinkedIn

Stable Diffusion has editing in the form of “inpainting” but the UI is still rudimentary. The decision is, should I render everything at home on my own hardware, should I spin up a GPU on Vultr, or should I pay for one of these services? And the value of the images needs to justify the time/cost of generating the images.

Seems like a powerful editing feature could justify the cost, but that advantage might be temporary as open source solutions tend to win market share in the long run.

The way I see this going, you’re going to see specialists, and AI models trained to do very specific subjects, and those specialists will build or buy those models to improve their work. Like with birds, what are the various feathers called, and what are the parts of those feathers? Are they fluffy or tapered?

Does an AI model even understand those specific descriptive terms, or is it just guessing? To know for sure, and to get much better results, an “eagle artist” specialist might purchase an AI model that was trained specifically to understand the qualities and behaviors of the feathers of American eagles, and the relevant terminology and habitats related to American eagles.

That will take human-work. And that human-work will be in demand, even though it might shift some of the market away from older techniques. Although it could create more work for photographers, to get new kinds of photographs to train AI models.

And it still takes an artist to recognize what’s working and not working in a composition, and editing functions will make these AI tools even more useful. Again, that requires human-work. AI will not steal everyone’s job. Some artists will make the transition, some won’t, and some won’t need to, because new techniques rarely completely replace old techniques.

As most artists know, there is value in keeping your process secret, which makes your work unique and original. When people use these services, they could be exposing some secret sauce without realizing it, and Adobe (for example) will take advantage of that, of course.

The pain associated with hosting AI rendering engines yourself, or rendering on your own hardware — it will be challenging and perhaps foolish, given that the paid solutions are so easy. But going out on a limb, doing unique things in unreasonable ways, is how artists set themselves apart from the competition, to capture our attention and imagination.