AI for Photographers in 2022

What We Will Cover in this Article

The scope of this article is simply to discuss the current state of AI for Photographers. It is not a “how to”, that may come later once I’ve established which platform is the most suitable for me.

AI for Photographers in 2022 has come a long way. Two years ago I wrote a post called “Does AI Improve Photo Processing” and concluded that AI was not a revolution because it is built on previous knowledge. (AI is “trained” by analysing vast quantities of relevant data – in this case, images).

That was two years ago and I was talking about the adoption of AI by vendors such as Adobe, Luminar, DxO, Imagen and ON1. Software development moves so fast though and we are now looking at a completely different landscape with a number of new vendors appearing in the market and a truly startling leap forward in capability.

It seems as thought there are two streams running in parallel, one dominated by existing photography vendors and another dominated by AI vendors. It would be a mistake to examine one without acknowledging the other, however I’m going to focus largely on the AI vendors in this article, before trying to imagine the potential in bringing the two streams together.

I’m going to demonstrate a small part of the capability of AI platforms from MidJourney, StableDiffusion and Dall-E using the same idea for each one and compare the output with an original photograph that I took a couple of months ago and processed myself. The idea is an image of windmills battered by dark storms on a mountain ridge. This manifests as a prompt for the AI. “Windmills battered by dark storms on a mountain ridge”.

MidJourney

The first image is generated using MidJourney. The process is more or less the same for all platforms, the user provides a prompt to get the base image and then refine the prompt to further refine the image. It’s reasonably straightforward to get to an image such as this.

MidJourney

MidJourney is a little more complex than the others to use – the interface resides in a Discord server so you have to have a Discord account in order to get to the user interface. As a beginner, this places your image in a thread for newbies which is constantly refreshed, meaning your image is pushed up the screen and is easy to lose track of.

StableDiffusion

The second image was created using StableDiffusion with the same prompt.

StableDiffusion

StableDiffusion differs from the others in that there is a photoshop plugin that enables you to create inside photoshop which is essential if these tools are going to become useful for photographers.

Dall-E

The third is generated with Dall-E,

Dall-E

AI for Photographers

The final image is my own photograph with a little drama added in post processing.

AI for Photographers
Canon 5D + Adobe Lightroom + Photoshop

It does look to me as though the capability of the underlying AI engines is ready to become useful for image processing. What I’d like to see personally is the ability to take a picture such as the one immediately above and process it using commands.

Where is the line between photography and graphic art? A simple enough question and I’m not sure I honestly know the answer. There is no doubt that post processing a photograph makes it possible to get a very different result than simply getting it right in camera.

Some would argue that real photographers get it right in camera and further processing is the work of the devil. I disagree strongly with this view. As would Ansell Adams and Henri Cartier Bresson, both of whom had a lot to say about the processing of their images in the darkroom.

As a photographer, you photograph using RAW images – that image is designed to be further processed using digital tools. I could also argue that our brains “process” what we see, so post processing happens all the time.

Where Now for AI?

I’m sure that the vendors will come up with surprising new twists in the coming months. Will AI reside in the camera or exclusively in the post processing workflow for example.

From my own perspective, AI image generation has a few steps to take that would make it regularly useful for photographers. The plugin for Photoshop developed for StableDiffusion by Christian Cantrell is an example of one of those steps. Here’s a wishlist if anyone is listening!

  • Plugins for all Editing Applications
  • Hi Definition
  • Capability to resize and change aspect ratio
  • Ability to build on an existing image rather than simply generate a variant

What would building on an existing image look like? This is where the two streams come together. Where tooling meets AI capability. It’s impossible to exclude editor specific capability from this list, we already have panels that leverage the power of photoshop. We need those to be developed not replaced.

What we need from the Photography AI vendors is capability that can be integrated into existing workflows.

  • Apps that generate a “look” from my own editing style (Imagen has a basic version of this already). To be useful it would need to be able to generate multiple models so that a black and white conversion for example doesn’t become an integral component of a single look.
  • Add layers containing new objects other than skies eg. A flock of birds, sunbeams, shadows, rain
  • Generate backgrounds suitable i.e. with space, to place images in.
  • Generate print ready versions of images. Something we can do in a few steps, but would be good to fully automate.
  • Upscaling. This is massive. Currently Topaz seem to lead the pack in using AI for specific functions, GigaPixelAI is the best upscaling tool I’ve seen.

I’ve put together a gallery of my own experiments in AI. This is scratching the surface, what would you like to see happen in the space?

Conclusion

The lists above draw out some useful observations – currently the photography industry is producing tools that enhance the technical business of making photographs the best they can be. Topaz, DxO, Adobe and On1 are producing great tools that bring technical perfection tantalisingly close.

The AI industry is creating impressive capability that seems to make AI truly creative. The observation that AI can recommend a song but can’t create one is beginning to look obsolete. AI very obviously can create images and the platforms I’ve demonstrated here prove that beyond doubt.

Will AI make Photography redundant? It seems unlikely. Amongst the reasons I take photographs of the landscape are the enjoyment of a sense of wonder when faced with a glorious combination of land and light. More prosaic concerns such as fitness and skill acquisition play a part too and I’m not sure that sitting in front of a computer screen will ever replace what for me is the essence of photography, being out in the land and finding the shots.

AI offers much for creatives, both practically and creatively. In terms of practical applications, generating backgrounds and variants of advertisements is one area where I can see an immediate use for AI. Want a Christmas flavour of an existing Layout? Ask AI. Or spend half a day labouring over the image in Photoshop.

Creatively, AI is a blank canvas. The only limit is your imagination. Whether that has much to do with photography will come out in the mix as the use of the tools becomes more pervasive.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.