Is AI or Deep Learning killing the job-market for creatives?

Last modified date

Short version: kind of! – I believe partly YES!

With the appearance of ChatGPT, Midjourney (using discord). DALL-E-2 and Stable Diffusion (open source) Deep Learning (DL) (some are AI) concepts seem to take over some of the markets for creatives (other models are clearly specialized on specific markets: BioBERT, camemBERT, BioNeMo – ). Btw this collection will be quickly outdated since this tech is developing right now with speed of light.

There are discussions everywhere about what these algorithms can do and how they have been trained. Incorporating techniques like “Style Transfer”, “Upscaling” and “composing” with AI tools will totally change the market I believe at least in some areas. Some impressive examples from Midjourney are available here:

Midjourney User Community Highlights from march 2023.

In general: its important to understand that these algorithms only work with an incredible amount of training data. Although there are billions of images online available only few are free to be used as training images for a deep learning algorithm. Even more critical since some of the AI/DL engines are not provided as Open Source – so images are created and sold based on images from the net that were not licensed for that usage. Most of the pictures online are only licensed for a specific use and are not free to be used for other purposes and clearly not for (other) commercial applications. I believe this is a huge legal issue that might slow the success of these picture deep learning (or “AI”) algorithms. Same applies to ChatGPT since it was also originally trained on text from the internet. So technology again has been changing what is possible, but only by overriding some legal guard railing and ignoring safety fence structures for the society. It seems as if the societies are once again just reacting to technology and not acting in an anticipating and forecasting way to these technologies (Deep Learning is available since decades and AI research is also not new). Social Media platforms are clearly out of control and since some years threatening an educated process of informing people for democratic elections – leading to echo chambers of fake news and crazy wrong narratives that can (and have been) changing elections in some countries.

An AI (Deep Learning) generated image from Midjourney – fully digital constructed by the deep learning algorithm.

The image above was created with Midjourney in February 2023. This is just a simple example of what this deep learning algorithms can create and by now things have likely already changed. In my opinion similar to ChatGPT this is a watershed-moment for a lot of creatives and especially for those that do abstract photography, illustrations, compositing and image fusion, for all those that try to make a living from abstract structures or patterns in photography. This especially impacts all variants of abstract photography art. Just imagine you need a nice shot from a beech tree with two kids playing in front of the tree for a commercial ad. You also want some old fashioned cars in the background and some big Lions in this image. You would need to hire a photographer that hires some models and the models would sign a model release contract. You would need someone to organize the lions (I have no idea how this would work btw), you would phone for a day or so to get these specific oldtimer-cars and someone must bring these to the set that you selected. You thought about the light and figured it has to be morning light at 7AM and they all have to be there the day before or at least early in the morning since everything must be in place in time. All this costs incredible money, comes with risks (will the weather be ok? do they all make it to the specific date/time and to your specific location? your pressure: you have to deliver these shots: so better make a water tight plan how to arrange everything with backup technical equipment and plan B ideas if something/someone fails) … etc – you will likely have some sleepless nights thinking about what can go wrong and do you really thought about everything … all batteries loaded? is my assistant fit? can I also do it when the lions dont make it to the location? and than after weeks of preparing everything the shooting works out fine and in post processing you managed to really nail it the way you and the creative director envisioned it. Finally you deliver your images and you hope that the client finally is happy. If possible you even wanted to exceed expectations etc.

All this just comes with a lot of unknown risks and will create a lot of stress moments and is extremely costly. But if you do these commercials its a way to generate income and there are various photographers specialized in specific domains of these shootings.

All over soon! In the not so remote future this is not needed anymore. With Midjourney you can ask the algorithm to create various versions of this scenario since the Deep Learning algorithm can for a fraction of the costs generate hundreds of versions of this image idea and the PR section of the client that plans the ad could easily select one of these images out of hundreds and regenerate variants from the best variant again and again. At the moment they would likely combine the AI generated version in this scenario with some real backgrounds or objects but this is likely not even necessary anymore in the future.

Think about another scenario: a nature oriented journal needs some nice pictures from butterflies of a specific type – lets say Lycaenidea (Bläuling). To get these for editorial use you would ask some well known nature photographers (if you want something stunning and new) or look for photographs in an image database – lets say or AdobeStock. In you can also select AI generated images. But with “Stable Diffusion” you can roll your own. No need for shutterstock or a nature photographers. Just generate your own images and lets get 100 versions and than lets discuss with the image editor or the group that does the article what works best. You can transfer various situations to this scenarios. I believe, press photography, weddings, sports and events in general are save. They must create authentic representations of the reality. But the full stock-photography market and these photography artists will have a hard time soon. I don’t believe loosing the stock market for photographers is a big loss – I never wanted to work for stock agencies, makes no sense to work on this basis imo – but overall in the future the market for Photography in general gets much smaller and there will be only demand for photography that is impossible to digitally reproduce since it has to show reality in its exact shape, proportions and appearance – capturing reality (sports, journalism, reportage, social and media events, science applications (photogrammetry/remote sensing/cell biology/astro physics), forensic applications and medical applications, documentations of all kind.

Other applications are also effected (3d animations based on physical properties / reflectance – to render for example texture and specific surfaces as accurate as in reality – Deep Learning can do this and might change also the way 3d-Animation software will work in the future (the big players are Houdini Software and Cinema4D). This clearly has consequences for 3D-Animation artists and specialist.

Unmentioned is the sound / audio domain. We can expect the same shift here towards more automatically generated content.

Are these deep learning concept generating art! No not really if we believe art is made by humans and a creative process. If you look closer: a lot of art made by humans is inspired (and trained?) by other art or ideas from others – so given that deep learning could be very creative in combining techniques humans could in return receive also a lot inspiration from AI generated image concepts. Right now its hard to predict where all this will lead to but if we look at ChatGPT first consequences are clearly visible: we have to carefully examine where text originally came from and who was writing the text and the same applies to pictures/painting and graphic art.

And who made the art when it came from an DL algorithm? The user or the algorithm? how to deal with copyright issues in the future? I really don’t like the feeling that DL art counts as art but I can understand the logic since some images from Midjourney are impressive. … BUT sometimes I have the feeling that this image was invented by someone else and Midjourney just learned that this description fits to this trained image – after all DL-based image creation could be in the end the biggest steal of creative properties in history.

And by the way – most of these algorithms use mainly deep learning technology (DL) and not AI but DL is just a small part of AI:


More to come – I guess I will somehow also try to use this technology but I am also deeply a believer of true photography. Finally photography only blows my mind when I can believe that it shows a variant of nature that really existed in space and time – otherwise … isnt it just an impressive painting done by a regression algorithm?