I believed AI would change Photoshop. Here is why I nonetheless do it myself

The obvious weaknesses of generative AI

One time, I used to be making a personality portraint for one among my gamers in an upcoming Alien RPG tabletop recreation. I wished a sci-fi man in a jumpsuit with company vibes and to have his fingers in “W” and “Y” shapes to symbolize his loyalty to Weyland-Yutani Company (his employer).

I struggled loads with that one. I imply, I simply wished the character to have his fingers splayed in the suitable approach. However may AI do it? Oh boy, may it not. Regardless of how arduous I attempted to finagle it, the outcomes sucked.

I attempted upwards of 10 completely different prompts to get it to know that I wished three fingers up on one hand and two on the opposite, splayed to create the impressions of “W” and “Y” letters. Typically it made the arms face the fallacious approach. It by no means bought the variety of fingers proper, and it by no means splayed them in the suitable approach. It completely failed.

After making an attempt—and failing—to get the AI producing what I wished for shut to twenty minutes, I gave up and simply made it myself utilizing the primary generated picture as a reference. I cloned one of many fingers, moved into the suitable spot, adjusted the lighting, blended the layers, and it got here out nice. All of that took 5 minutes.

Okay, advantageous. You would possibly say that I constructed upon the unique creation put forth by the AI. Sure, I’m glad it gave me that preliminary design to work with, and it was simpler to edit that than create the identical factor from scratch. However as a remaining product? It did not do what I wanted. It was no extra usable than a random picture I may’ve grabbed from someplace on-line.

Most of the time, it’s simply faster for me to make edits manually. With AI, I’ve to consider how one can immediate the AI appropriately and ensuring the AI will interpret my directions the precise approach I intend to get what I want from it—not one thing else, not one thing that will get it proper in a single space however randomly modifications one thing in one other space, not the suitable picture however with an overhauled artwork type. With the most recent crop of LLMs, doing this proves frustratingly and opaquely tough.

And that’s my most important downside with right this moment’s generative AI: it takes a lot time to repair what it creates that I’d as properly have executed it myself within the first place. Simply have a look at the notorious Coca-Cola Christmas advert that got here out atrocious but price way more to create than in the event that they’d simply paid some animators and artists—and so they ended up doing that anyway as a result of the AI outcomes sucked and wanted to be punched up.

Now, in case you’re a person and also you don’t know the very first thing about Photoshop, then AI could be a software to get you midway there. However we’re not but on the level the place guide changes, punch-ups, and handcrafted artwork—even by amateurs like myself—are out of date.