Experiments I conducted with DALL·E 2 from @OpenAI replicating styles of well known portrait photographers using photo-realistic AI. 🧵
1. Dorothea Lange
2. Helmut Newton
In each instance I used the phrase “Photo of a young Mexican woman in the style of [photographer]”
3. Diane Arbus
I chose this search phrase as I am currently living in Mexico City, but it is fun to imagine how classic portrait artists might have captured other cultures and ethnicities.
4. Eric Lafforgue @ericlafforgue
DALL-E has an incredibly refined understanding of lighting and composition.
5. Annie Leibovitz
For each artist I generally completed 2 or 3 iterations, with each iteration returning six unique images. I then chose the best 4.
6. Lee Jeffries @Lee_Jeffries
When the source artist has a very consistent style, like Jeffries here, DALL·E is able to replicate it with impressive accuracy in some cases.
7. Steve McCurry @McCurryStudios
The results are less impressive where the source photographer (in this case my personal favourite) is known for diverse work that often ties the subject with the environment.
8. Nancy Goldin @crazynangoldin
While this is most definitely art, it is not yet known what impact this will have on the future of photography.
9. Robert Mapplethorpe
In this Mapplethorpe series, DALL·E, for many of the iterations I did, would combine his work with flowers with his portrait work.
10. David Bailey
In many of these renders you can immediately recognise the source artist. If it were not AI you would accuse the photographer of style plagiarism.
11. Terry Richardson
DALL·E has pretty good command over the use of clothing accessories and expressions. It knows what attitude Terry captures in his work.
12. Yousuf Karsh @KarshYousuf
The exact status of copyright is too complicated to put into a tweet. But simply put, I do not own the copyright to the digital file (yet).
13. Martin Schoeller
Again here, for artists like Schoeller with an extremely consistent visual style, DALL·E is able to replicate with ease.
14. Julia Margaret Cameron
Even when the compositions vary you can still see very strongly the inspiration of the source artist.
15. George Hurrell
I don’t understand how any of this works.
16. Peter Lindbergh @peterlindbergh
One thing that impresses me is the ability to capture the gaze and match that with the source artist.
17. Brian Ingram
I’m really most impressed with how often it gets everything right whilst pulling something creative out of nowhere (but not really nowhere!).
18. Kenneth Willardt @KennethWillardt
Although some images have minor glitches, these areas can be re-rendered by DALL·E using a brush tool. In my experiments I have generally kept the original first render.
19. Richard Avedon
It appears there might be content filters AFTER the images are generated that scan for nudity. My search on Sally Mann was rejected.
20. Dmitry Ageev @ageevphoto
Dmitry is an active NFT artist. One thing @OpenAI very clearly state is that rendered images cannot (currently) be sold as NFTs.
21. Rosie Matheson @ Rosieyeah
DALL·E appears to be effortlessly creative. Every time you repeat a search description it seeds a different set of results.
22. Herb Ritts
For this experiment I chose a few well known photographers, a few with distinctive styles, and others that I am just a fan of.
23. Arnold Newman
It has the most difficulty with eyes - a critical aspect of portrait photography. A lot of glitches, but then sometimes it renders with remarkable detail.
24. Robert Frank
I generally stuck with the same search phrase but more accurate results could have been achieved with a more detailed description.
25. Bruce Gilden
One things these images all have in common is that they are quite tightly cropped, regardless of the style of the source artist. For some I added ‘view from a distance’ to zoom out but results were mixed.
I did try a DALL·E render for Vivian Maier but every single result came up with a freaky camera (not even her famous Hasselblad) so I guess her work is forever going to be linked with the selfie. I even tried adding “without camera” but it wasn’t smart enough to understand.
Although DALL·E works exceptionally well for some photographers, it really struggles with others. These Gregory Crewdson shots are thematically correct but details of figures are all really glitchy. @crewdsonstudio
If you want to learn more, @MKBHD has a great video on this tool I appreciate the opportunity to have an early look at this technology. Thanks @OpenAI
Share this Scrolly Tale with your friends.
A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.