ChatGPT and other AI platforms can produce some amusing, disturbing, and even adorable (if often in a possibly copyright-infringing way) images. But, much like how the AI tends to default to an overly optimistic and encouraging tone in its written responses, there’s a tendency to soften or flatten image styles without explicit instructions, even when the source is as quirky and unique as, say, a young child’s doodles.
You can overcome ChatGPT’s preference for generic recreation of drawings with a little extra effort. Several people have shared their successes in bringing their children’s drawings to “life,” without ChatGPT turning a silly drawing of a bird into just another robin you’d see in your yard. Based on those and after some experiments, aided by my young son and his enthusiastic scribbles, I came up with a pretty successful, though still inconsistent, template for accurately bringing his interpretations of a dog, a bear, and a dragon a simulacrum of reality. The prompt is not short, but to work around ChatGPT’s preferences, it’s all necessary and might even benefit from being longer:
Turn this child’s drawing into a photorealistic biological creature, using it as a strict blueprint: preserve the exact silhouette, proportions, feature placement, asymmetry, and color intent so the result would align if overlaid. Convert lines into natural anatomical edges, flat shapes into volume within the same boundaries, simple eyes into realistic wet eyes in the same position and size, mouths into subtle skin creases, and stick limbs into thin but believable anatomy. Translate the original colors into natural pigmentation within the same family using tonal variation only. Add realistic surface detail such as fur or skin texture, gentle folds, small imperfections, natural color variation, and eye reflections. Present it as a serious wildlife photograph of a believable species, shot with an 85mm lens, shallow depth of field, natural lighting, high detail, and a simple natural background so the creature remains the focus.
Article continues below
Chihuahua toy
It’s worth mentioning that the drawings were the result of a lot of negotiation with my child, whose initial attempts were, charitably, from the surreal impressionist school of art. And to keep it simple, we stuck to black and white, but the AI understood well enough what was being requested to come up with colors.
My kid’s attempt to draw a chihuahua wasn’t too far off base, though the giant head and uneven stance suggested a creature in mid-wobble. The photorealistic version did not correct a single one of those issues. Instead, it leaned into them. The oversized ears became finely detailed. The eyes, once simple circles, turned glossy and alert. The tiny body and slightly off-balance posture remained exactly as drawn, which gave the whole image a strange credibility.
What stood out most was how natural it looked despite everything being wrong. The proportions were still exaggerated, but the realism made them feel like a deliberate quirk of the animal rather than a mistake. It looked like a real creature that had simply evolved in a slightly different direction.
My son’s reaction was immediate recognition. That, more than anything, felt like the point.
Bear in the woods
The bear started as the simplest of the three drawings. No matter how many times the drawing started as an adult bear, the resulting drawings were closer to a Teddy than a grizzly.
ChatGPT’s interpretation looked halfway between a Paddington doll and an actual bear. The head came out somewhat real-looking, though as a much more intelligent creature. The rest of it ended up more magical soft toy brought to life than a natural beast.
Still, though the proportions were those of a child’s imagination, the realism gave them weight. My son still saw it as cuddly. From a distance, so did I. Up close, it felt like it might have its own opinions.
Dragonfire
Without deliberately influencing his subject choice, I might have been hard-pressed to realize my son was trying to draw a dragon over and over. While letting him take the lead as much as possible, I did a little extra nudging to produce a dragon on paper.
Of course, dragons were never really in a natural landscape looking like actual biological species, but this virtual one checked off every box for my son. Personally, I found this interpretation a little creepy. Uneven wings and legs, and a spine curved in ways that felt uncomfortable to observe. But, as that was basically what I’d asked for, I appreciated the effort. And at least the flames looked cool.
If you’re wondering why I might have spent so much effort on the dragon in particular, it was mostly to try to erase the memory of ChatGPT’s attempt to produce a hypothetically real version of my son’s artistic interpretation. As you can see below, there are things much more terrifying than a slightly off-kilter dragon overhead.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.
The best business laptops for all budgets
https://cdn.mos.cms.futurecdn.net/LQUZrtv3TKcaMMW9qHfJr8-1920-80.png
Source link
ESchwartzwrites@gmail.com (Eric Hal Schwartz)
