Image generation, photographic competitions and democratising creativity

The development of systems, such as Dalle-E 2 and Midjourney, that can generate photo-realistic images that are practically indistinguishable from images captured using a digital camera has, understandably, raised a number of issues for amateur photographers and, particularly, members of photographic societies and camera clubs.

Generative image systems are machine-learning systems built around a huge corpus of images that have been scraped from many different sites around the world. They generate images in response to text prompts such as ‘create an image of fishing boats in a small harbour at sunset with mountains in the background’. These images can replicate images created with a digital camera and at least one such generated image has won an award in an international photographic competition.

The situation is confused, in my opinion, by the extensive use of the terms ‘AI-based’ systems which, technically, doesn’t really mean very much. Computational photography has been around since the 1990s and the technology has improved dramatically in that time. The latest technology iteration is based around so-called machine learning systems but, in many applications, these simplify automate manipulations that were possible with previous technology.

Competitions are the lifeblood of camera clubs where members enter their image and compare them with those of other members. Competitions are a stimulus for members to take photographs and, with informed commentary, help them improve their photography. Arguably, however, competitions encourage conformity and may sometimes stifle creativity where images do not conform to what judges consider to be a ‘good’ photography.

International photographic societies such as the Royal Photographic Society in the UK are debating how to respond to these image generators and have issued some preliminary guidance on the use of generative systems. They also recognise there are no simple answers here and that much more discussion is required.

There are three principal scenarios of use of generative image systems:

  1. To generate images that look like photographs that could have been captured using a lens-based system. An example of that would be the fishing boats in harbour at sunset that I suggested above.
  2. To add generated elements to an existing photograph captured using a lens-based system. For example, a dramatic sky with towering cumulus clouds could be generated and added to a landscape photograph.
  3. To create ‘unreal’ images which could not normally be created using a lens-based system. I use the term ‘normally’ here deliberately because the capabilities of photo-editors such as Photoshop are practically unlimited. However, the creation is extremely complex composite images is beyond the capabilities of the vast majority of photographers.

From a photographic competition perspective, it is fairly easy to take positions on the first and third scenarios. In my opinion, photographic competitions should be restricted to images captured using a lens-based system. Real or unreal images that are wholly computer-generated should be excluded. It may well be that some photographers are interested in exploring generative systems as a new form of digital art and this is perfectly valid - but they have no place in photographic competitions.

The middle option above is more contentious. Currently, those with Photoshop skills can create remarkable composite images that combine parts of several images, often manipulating some of these parts by changing colour, size and so on. Is this very different from creating component images using an image generator and combining these with the aid of the generation system? The counter to this, of course, is that competitive use of composite images requires all parts to be the copyright of the photographer but I don’t believe that the current rules forbid some of these parts (eg textures) being produced by a system that is not lens-based. In reality, image generators will surely allow, for a fee, copyright images to be generated so a photographer could, for example, create a library of dramatic skies and legitimately include these.

Image generators offer the possibility to simplify creating composite images. Arguably then, these generators might be seen as democratising creativity so that those with creative ideas do not have to spend a great deal of time becoming proficient in advanced features of Photoshop.

What does this mean then for photographic competitions? In my view, we need to create a separate class of competition for composite images. There should be a requirement, perhaps, that these should consist of a principal image that is created using a lens-based system with composites that may be other images taken by the photographer or generated images or a mixture of both. Instead of being judged against ‘conventional images’, composites are therefore judged against each other. The other class should be single images in their entirety, taken with a lens-based system.

Of course, an argument that will be raised will be that ‘composites for improvement’ are different from ‘composites for creativity’. An example of a composite for improvement might be to change the sky on a landscape image to something more dramatic. It may well be possible to tweak the rules to allow this but I’m not personally sympathetic. As soon as something is allowed, we enter a murky area of where to draw this line. I prefer the simplicity of ‘pure’ images with no composite components.

Clearly this whole area is evolving rapidly but, in my view, we should not demonise image generation systems. They actually offer an opportunity to rethink photographic competitions with, perhaps, more support for creativity and so include many photographers whose work does not fit the sometimes rigid notions of what makes a ‘good’ competition photograph.