How AI-Art Could Inspire Portrait Photography
Updated: Oct 12, 2022
Use Case I - Post-edit of a Portrait Shot
When I first heard about AI text-to-image engines I found that quite an interesting new method to create any kind of images. But images created by a computer - is that a competition to photography? Is it "art"? Is it "creativity" when using it? Is AI-based automatic replacement of the sky in a landscape photo something a landscape photographer should use? To which extent should the AI-based smart optimisation of a portrait or body shot be applied to a human face or body?
All those and similar questions are only one and the same question in the end and was asked since the beginning of photography: How far may image manipulation go?
Now with AI-based image generators like Midjourney, Stable Diffusion or Dall-E 2 this question came up to me again. But rather than asking the primal question again, I thought about how I could use AI art generators to create new ideas for people and portrait photography. I registered at Dall-E 2, where you have 50 so called "prompts" (textual expressions) for free. After the quick read of some tutorials on the how-to of effective prompting I started right from scratch with my first experiments. I had several use cases for potential future photo projects in mind, and the first one is subject to this post:
Changing or Adding Costumes and Background of a Portrait Shot in Post
Dall-E 2 offers a mode to "out-paint" new content to a given part of an image according to the prompted imagination of the user/photographer.
I chose the picture below of Vanessa that we shot in early 2022 to change everything around her face and hair.
In a first step, you have to upload the original image to Dall-E and crop it to square size. Then you erase everything in the image that should be replaced by the prompted content. The eraser tool works quite rough. Later I found out, that there would be more precise results when the original image gets cut out properly in an editing software and is saved as png with transparent background.
Now comes the experimental part: I wanted some mystical red dress and some fantasy hair dress to blend with the portrait. After some trial and error I ended up with this prompt:
"dslr studio shot of female model with mystic headdress, in award-winning costume design of woman’s red dress inspired by "LORD OF THE RINGS", misty background, from Lindsay Adler"
Now, here are two of my favourite results (after some final retouch):
How about portrait variants of a western-styled lady , a combat pilot or a space ship commander? After some prompting trials and additional edits (like lens flares) I had those results:
After my first experience with Dall-E 2 - what is your opinion about the generated portraits? Who is the creator, the AI or the photographer, or both of them? Or should we rather say, that the AI-software is just another creative tool like the renown image editing apps with all their filters, layers, background and foreground manipulation, light, atmosphere and color changes - but of a more creative and unpredictable kind? Let's do some analysis of what we have created here:
The original photography (the portrait) was created by me, the photographer. I made the shot, I used studio flash and background lighting, went through the whole post processing - starting from image selection, through RAW development, beauty retouch and final color grading. The model did a great job and we have a real existing human face and expression in the image. I as the photographer own 100% of the copyright.
What I prompted as a description to aim for, is my creative job and work, which needs my skills in precise writing and vocabulary. The AI didn't do that by itself.
However, some of the prompt terms refer to the creative work of others: I used e.g. "Lord of the Rings" and the name of the fashion photographer Lindsay Adler to refer to her portrait style. Now, the AI-art generator combines elements of both. This is not copying the work of others, it is just miming their style or typical design elements. Everyone is allowed to do that, e.g. may an artist paint an oil painting in the style of Rembrandt (van Rijn). Only in case of a perfect copy of a Rembrandt painting claimed as an original "Rembrandt" for sale we would say this is a criminal act. But if the painter or AI would just generate an alternative version of an original Rembrandt painting, everything is legal.
What is really amazing is, that Dall-E 2 does a perfect matching of the new image content with the given one. The extension and blend of the hair, the color and direction of light, the color palette for costume and background, the camera perspective and the depth of field. I guess, that the future will show us even more precise results (e.g. image resolution, blends, detailing).
The final work created is a unique workpiece. The AI-art generator is not a real person or a person in legal sense - as it says, it is only an artificial instance. Therefore, the "AI" cannot own any copyright and thus it belongs to the "creator", who formulated the prompt according to her/his intent (in this case the photographer). No single element in the image can be found in any informational resource like internet, magazines, books or the like. The created outfits can't be found anywhere on the planet or in the universe, no copyright claims will be expected. There is no reference to any external image resource that would have to be quoted next to the picture.
An open issue is about the inclusion of a real model. The AI-art generators basically refuse to do any manipulation on a human face or accept sex, crime and violence oriented requirements. However, as we can see in the examples above, the surrounding and context of a model portrait can be manipulated. In a so called model release - the contract between model and photographer on the image use and rights - it is usually agreed between both parties that an image may not be manipulated in a way that would be disadvantageous or harmful to the model or photographer and/or her/his reputation. In any case, publication of a strongly modified image must be agreed upon by the model or the photographer (dependent on who of them made the manipulations of the work). For this series of posts, the consent was obtained.
In the upcoming post (Use Case II) we will prompt Dall-E 2 to create a mood board for a portrait shooting.
All chapters of the series "How AI-art Could Inspire Portrait Photography":
Use Case I - Post-edit of a Portrait Shot
Use Case II - Creating a Mood Board
Use Case III - Creating Outfits
Use Case IV - Creating Background Images from Scratch