First things first, I need to conduct some business. This blog is something that I really want to grow an audience for. If you are reading it and enjoying it, please help support it by joining my mailing list using the form at the bottom of this post. Okay, now that’s out of the way!
I wanted to try an experiment using some classic databending techniques combined with some of Midjourney’s features to try and create a new aesthetic. My plan is to first generate a clean image from Midjourney, convert it to .bmp, open it as raw data in Audacity, and to manipulate the data with effects and editing. Then, I will use Midjourney’s “Describe” feature to have it generate a prompt it thinks would give me a similar image. I will then iterate through the results until I find what I want. Once upscaled, I will use Midjourney’s “Blend” feature to blend it with the original output from Audacity.
Step 1: Getting a Clean Starting Image
The first thing I need to do is determine the subject matter I want the final piece to represent. In this case, I want there to be two crows sitting on a power line next to each other. Next I need to determine what I want the final aspect ratio of the piece to be. For the given subject matter, I feel like a 4:3 aspect ratio will be good. So now it is time to do some prompt engineering.
My first prompt into Midjourney keeps it super simple:
/imagine 2 crows on a power line next to each other –ar 4:3 –q 2 –v 5
Midjourney pretty much gave me exactly what I asked for, but it isn’t looking exactly how I want it. So, let’s add some details to the prompt and see what we get. I will start the prompt with “Darkness broods…” in order to add some drama and moodiness. Added some keywords like “photorealistic” and “wide angle” to give more style to the image. –no dof removes depth of field. Let’s try again:
/imagine Darkness broods while two crows sit next to each other on a power line in an ominous scene. photorealistic. wide angle. cgsociety. hyperrealistic. –no dof :: cinematic :: realistic :: intricate :: detailed :: 6k :: –aspect 4:3 –q 2 –v 5
That is looking way better. Next, I will re-roll the output in Midjourney and iterate through to see if I get anything more interesting. Here are some iterations:
Alright, I really love the upper right corner image in the middle iteration. The negative space between the crows is beautiful. So, I upscaled and saved the image as a PNG from Midjourney:
Step 2: Manipulating Data with Audacity
Next, I saved the image as a .bmp file. Normally when I databend, I use non-interleaved TIFF files because the quality and color control is better. Because of the minimal color palette in this image, I think .bmp will actually be better. Once I have a BMP, I open it in Audacity as raw data:
Now, I select everything after the first 3 seconds of data in order to make sure the header is still intact. I applied a bass reduction, treble increase, and slight volume decrease. I then selected a random chunk of data in the middle and applied an echo effect. I selected another random chunk of data that slightly overlaps the previous section and applied a reverb effect. I encourage you to experiment and play around with different effects and settings. It’s crazy what you can do with raw data in Audacity! So now I export the file and here is what we have:
Woah! This is not what I was looking for, but I love it, so we’re going with it. That’s one of my favorite things about glitch art. Some of the best outputs will be accidents.
Step 3: Have MidJourney “Describe” The Audacity Output
Next, we will ask Midjourney what prompts it thinks would generate an image like this using the Midjourney “Describe” feature. Here are the prompts it suggested along with an output for each:
/imagine an abstract of birds perched on a wire, in the style of glitch art, romantic graffiti, made of liquid metal, thermal camera, colorized –ar 77:58 –q 2 –v 5
/imagine a colored digital painting of birds, in the style of romantic graffiti, made of liquid metal, photo taken with ektachrome, light black and indigo, richly colored skies, two dimensional –ar 77:58 –q 2 –v 5
/imagine two birds on a wire, in the style of glitch art, cosmic graffiti, hdr, spray painted, thermal camera, expressionist: emotional intensity –ar 77:58 –q 2 –v 5
/imagine two birds are sitting on a branch near a field of grass, in the style of glitch art, street art photography, expressionist: emotional intensity, made of liquid metal, dark, foreboding colors, romantic graffiti, orientalist imagery –ar 77:58 –q 2 –v 5
A lot of these look pretty cool! Here’s my favorite output upscaled:
Step 4: Blending Midjourney’s Description Output with the Audacity Output
Next, I will use Midjourney’s “Blend” feature to blend this image with the databent image we got from Audacity. Here is my favorite output after blending:
I do like it, but if I’m being honest I prefer the output we got from Audacity. Midjourney’s outputs have lost almost all of the databending aesthetic. Fear not! We can do several things to remedy that. One option would be to blend this output with the databent output from Audacity and repeat that process until you’ve forced your will on the AI. I tried both options. Here’s the output Midjourney gave me after several repetitions of re-blending it with the databent file from Audacity:
I definitely am digging this one! Here’s one where I took Midjourney’s blended output and did another round of databending in Audacity:
I love it! Okay, so one of the best things about glitch art, generative art, AI art, or anything else that is iterative like this is that any given step can produce art that you love. I love the aesthetic so much I will probably create a series from this.
Make sure to experiment. Find the limits of what you’re working with. Push the effects in Audacity until you get no output so you can familiarize yourself with what you can and can’t do. Never stop breaking things!