Bootlegs: Turning Photoshop Fails into Fashion
New tools, new glitches, new gateways to innovation
Hater’s gonna say it’s Photoshop.
Every new digital tool shapes our design process with unique limitations and unexpected quirks.
Adobe's push into Generative Fill is a fascinating study of potential versus performance. As the tool evolves in Photoshop Beta, it has room to grow, especially when compared to the dazzling capabilities of popular online image generators.
However, it’s in these gaps that accentuate technological imperfections that one can find unexpected sources of inspiration.
Process
I used two PNG sprites from the Mario Kart DS game. I created a simple composition with the character faces staggered and then used a series of simple prompts to have Photoshop create a series of reimagined Mario Kart universes and scenarios.
I had no idea what was going to happen at first, but I knew I wanted this:
A dramatic shot where the characters are in front of vehicles.
The Mario character had to be in the foreground to show he was the leader.
The image needed to be isolated to be easy to cut out and use further.
Finally, one prompt provided a series of acceptable images, which I refined further to produce more similar compositions. The prompt was:
Mario and Yoshi from the Mario Kart series on Nintendo DS. Yoshi is on the left and he resembles a dinosaur. The character is Mario from the Nintendo games. He is dressed in blue overalls, a red shirt underneath, and black boots. They are standing next to their race cars at a pit stop. The background should be vibrant colorful and fun like the Nintendo DS Mario Kart game.
Nothing in these images was “good” at first, but with some adjustments and retouching, a decent composition eventually emerged.
Using Stable Diffusion and Krea.ai for Reintorpolation
Photoshop’s Generative Fill and Firefly as a model are nowhere as useful as something like Midjourney for creating quality source material with just a simple prompt. It probably does not help that Adobe's floating prompter is a nuisance to visual workflow.
However, knowing that I was going to process this image further using a separate model, all I had to do was make sure the source image is closer to the final composition that is emerging.
Generative fill is great for rough retouching. First, the figures and the vehicles were properly isolated from the background, and I used a few simple prompts to remove the odd golden gloves and replace them with workwear, along with a few other minor things that I did not want to influence the next model.
The image was then imported into Stable Diffusion img2img and Krea.ai to upscale and enhance further. After several rounds of prompts, the final, more highly detailed style was produced below:
At this stage, I personally found this extremely novel and funny but useless. I felt like I had made some pixelated characters in a different style but not much else.
The answer, of course, was to reinterpret them further and bring the new characters back into the source material's original universe. I used Photoshop to create custom bootleg text and adapted the hyperrealistic CGI image style to an accurate pixelated, 16-bit aesthetic.
I wanted to place our new Kartbros in the action by creating a dynamic, manga-inspired poster. I gave the image above to ChatGPT and worked through a series of prompts to produce a series of pictures to create a story. Given that these would be stylized much further in Photoshop for the final composition, the accuracy or quality of these images did not matter much.
By letting go of control, designers can turn imperfections, faults, and limitations into profound creative opportunities. Embrace the imperfections and let them guide you to new and exceptional world views.
See you in the Super Kartbros Universe!