10 Photos Created By AI That Don’t Exist In Real Life, As Shared By This Digital Artist
A photographer and digital artist Aurel Manea has created some beautiful landscape photos with the help of the new Stability AI. An open-source text-to-image AI tool called Stable Diffusion helps produce images based on the descriptions provided.
In an interview with Demilked, Manea said, “I was a landscape photographer all my life, I worked hard for my photos and yet I wanted to see how this turns out. So I just added the name of my favorite landscape photographer (Marc Adamus), some modifiers like dramatic lighting, sunset, beautiful, landscape photography”+the description of what the scenes were. I was just blown away by the results so I had to share them.” Check out some of his works and the rest of the interviews in the gallery below.
#1
Image source: Stable Diffusion
The artist started creating AI images around 3-4 months ago with an AI called Disco Diffusion. He told DeMilked that he had to wait 40 minutes for a single image to generate. According to him, “it produced very distorted, very incoherent but still quite beautiful. He worked with several AI software and found that “most image generating AI’s give beautiful images but a lot of times broken apart, multiple heads, hands coming out of the feet, etc.”
#2
Image source: Stable Diffusion
Manea then discovered another software called Midjourney before it went public. He says, “I was simply amazed at what you could achieve with it and you can see that from all the images popping up everywhere. They now have more than a million users and I think they are by far the most used. The problem with Midjourney I guess is that it can’t do photographic images and landscapes and it has a visual fingerprint. You can tell instantly this is an image generated with Midjourney. A bit grungy, a bit brown, very strange high-level artefacts, etc. But for me, the generation times compared to Disco Diffusion felt like a miracle.”
#3
Image source: Stable Diffusion
After exploring several AI software, Manea saw the link from Stability AI that they wanted testers. He explained, “I looked it up a bit, very few images but showed a lot of potential. And after 3 weeks of waitinng I got in, I got my discord server to invite and started “prompting”. 9 images in 10 seconds (it is now only 4 images possible at a time), incredible flexibility regarding visual style, incredibly beautiful outputs (with the right words), etc. I was beyond impressed. It was also beautiful to see what other people were creating, stealing some parts of their prompts, improving upon them. In a few days the general beauty of what people created improved drastically.”
#4
Image source: Stable Diffusion
He adds, “In the end, I also got my Dalle 2 invite but was disappointed. Generally, it can produce beautiful photo-like images of people, animals, and things (better than Stable Diffusion) but it does not have the explorative factor of Stable and styles. And the absurd pricing and censorship just don’t encourage exploration.”
#5
Image source: Stable Diffusion
The artist also shared the creative process with us- “Think of something that you want to have materialized into a beautiful image, try to describe it as good as you can (In AI language, in so many ways is differed from ours and you run into impossibilities all the time). Try to reference artists that feel right for what you are trying to create, mix them together if you like. Try using parts of the words other people used (I keep a huge list of prompts + resulting images), fit them into your idea. Not good? Try again, change the prompt, move words around to give higher priority, remove words and styles that might conflict with your original idea (Marc Adamus alien landscape simply does not work for example and it is useful to use artists that are known for images that are in a similar universe like the one you are trying to build). If I like the images I save them, upscale them using Gigapixel AI, share them.”
#6
Image source: Stable Diffusion
Manea also shared some disadvantages of using this AI – “I do feel a bit guilty about these images. Most of my friends and followers are photographers, I think most of them feel threatened or just disappointed in me. I lost a lot of reach and I think a lot of people unfollowed me. I guess it is only natural when new tech comes along, there is a clash of cultures but I think in the end all of us will express ourselves in images. And like Emad, the CEO of stability said it is better to create than to consume.”
#7
Image source: Stable Diffusion
In the end, Manea added, “Generating images even if they are very satisfying they feel a bit like fast food. I wish I will keep doing photography as well. If not for the result but for the journey.”
#8
Image source: Stable Diffusion
#9
Image source: Stable Diffusion
#10
Image source: Stable Diffusion
Got wisdom to pour?