THE WORST Ai COSPLAY VOL. 1

Project: Make “the worst” Cosplayers at a convention portraying Megaman 2 Robot Masters.

Experiment: How accurately does Google’s Nano Banana translate abstract imagery (Megaman 2 pixelart) into photorealism? How expressively does Midjourney translate static imagery to motion?

Tools: Google Nano Banana (Image to Image), Midjourney (Image to Video), DaVinci (editing). 8 clips total. 

Vision: A diverse array of people awkwardly but joyously presenting their homemade costumes at a convention. Intro of the Robots from the game with a pause so you can compare the costumes. Emphasis on humanity, silliness and expressiveness. 

Expectations: Prior to this project, I had never tried to translate one image to another using Gen Ai. Admittedly, I could only theorize on how to pull it off – and improvised my pipeline as I was building. Conversely, I had full confidence Midjourney’s Image-to-Video would create excellent animatics that easily sailed past the uncanny valley. So I expected to struggle with Image-to-Image and breeze past Image-to-Video.

What actually happened: I was genuinely blown away by the results for both the imagery and animation. It took a little bit of practice to translate the sprites (the people in cardboard are my first attempts) into something more believably “homebrew” – but I was impressed how Nano Banana translated colors, designs and some motifs from the pixel art – though I had a pretty active hand in the costuming once I got the hang of it. 

What I didn’t expect: It’s the animations in Midjourney that really sold it for me. There’s a life, dynamism and joyousness that I feel translates through the screen. The first guy even seems to mouth “Aw man!”, which I don’t think on my best day I could ever recreate. That “Aw man!” also brings up an Uncanny thought: these people aren’t real, but the warmth I feel when I look at this goofiness is authentic. 

Results: Yes, you can translate fairly abstract imagery (in this case pixelart) into photorealistic equivalents, it just takes conscientious human oversight. And yes, near as I can figure, you can coax an unfeeling machine to evoke vulnerability and joy. As with most things Ai, the missing element is Humanity. 

🔴If You’re Anti Ai: The results of this experiment were still wild and unpredictable. You can successfully position yourself as offering far more control over scenework and acting. Ai is still time-consuming and produces waste clips at a distressing ratio Larger projects with greater expectations can not succeed off of roulette wheel spins – and that’s what Ai feels like on a project.

🔵If You’re Pro Ai: You absolutely can move past Ai Slop by focusing less on execution and more on concepts and emotion. Think about what you want your audience to feel, then work backwards. You’ll save yourself dozens of Fast Hours / Credits / Sanity. 

All the source files can be found here so you can experiment yourself (right click save as).