My Comics Collaboration With DALL-E 

The primary immediate I typed into DALL-E 2 was “Cubist portray of a tuxedo cat taking part in piano.” The AI tinkered and buffered for a minute after which delivered 4 completely cute pictures of cats taking part in piano. The shock, whimsy, and shade impressed me — in spite of everything, many people have seen a number of the extra disturbing outcomes of AI-generated pictures. 

I’ve been instructing and writing and dabbling with comics for a few years now, so it appeared pure to experiment with DALL-E 2 to make a text-image collaboration. Might DALL-E 2 and I develop a shared, distinctive “voice” in our artistic output? 

Sometimes, I make comics in one in every of two methods: I draw and write one thing shortly in a diary or I spend months scripting, writing and rewriting, drawing and erasing, scanning and modifying a half dozen pages. Making comics is a laborious course of, particularly for single-creator comics. On this “auteur” mannequin of creating comics, the one that writes the comedian additionally illustrates it. Stemming from underground comix (transgressive comics that didn’t meet the approval of the Comics Code Authority) and different comics (different to, say, mainstream Marvel and DC comics), these hand-drawn comics mirror the distinctive, developed fashion of their creator. From the gorgeous brushwork of Craig Thompson to John Porcellino’s clear traces and distilled pictures, a comic book creator’s panels grow to be as recognizable as a author’s voice or an artist’s fashion.

Cartoonist Lynda Barry famously champions the fantastic thing about the hand-drawn line and the individuality of its expression. Even in case you are “dangerous at drawing,” you make a picture distinctive to you; Barry would say your drawing is extra “alive” than the drawing of an expert draftsman. There may be, she contends, one thing private and actual about an inexact representational drawing. Even the bodily look of the textual content turns into one other methodology of communication, a form of picture, as a result of handwriting carries with it an innate expression of character and individuality, providing one other layer of tone and which means.

A few of DALL-E 2’s drawings have a few of this “dwelling” high quality. After I uploaded one in every of my very own sketches of a mischievous cat in a witch’s hat, DALL-E 2 made 4 variations that, to me, appeared like they might be “inhabited” characters. But it surely’s laborious to persistently replicate these qualities with a brand new immediate (to be truthful, that is additionally a problem for me). 

A full collaboration between two AIs to “generate” a comic book is feasible now that OpenAI has launched ChatGPT and InstructGPT, which generates writing when prompted. However comics depend on a poetic precision of line — there’s not a lot house for expository textual content, and I prefer to keep away from having the phrases duplicate the work the pictures are doing.

Describing the picture I wished the AI to ship took a distinct form of finesse and revision that I used to be accustomed to. The immediate field jogged my memory that longer, extra particular descriptions will give higher outcomes. Due to this fact the writing work is doubled, scripting the comedian and describing desired potential illustrations. “Chalk pastel drawing of an evening sky with a number of stars” proved a disappointment however generated a brand new thought: “Chalk pastel drawing of a Kentucky horse farm at night time with a number of stars within the sky.” Was I coaching the AI? Was the AI coaching me? Or did we have interaction in an iterative course of as collaborators? I ended the experiment, questioning what it would take to make a full collaboration during which the pictures complemented the writing and felt expressive, “alive,” fairly than soulless.