Page 9 - November Merged
P. 9
Feature | Stop using ChatGPT
BEST CHATBOT FOR GENERATING IMAGES
Our first image-generation test involved spatial awareness, colour and lighting. because AI often has trouble placing
creating an original photo from a detailed Our second test entailed generating a people behind objects, and two signs we
prompt. This was to gauge how accurately composite image by placing the subject of asked it to replace to test its text rendering.
the AI chatbot could combine the one photo on the background of another. Claude doesn’t generate images, so we
different elements in terms of realism, The second picture included a barred gate left it out of this round of testing.
Prompt 1: Generating an original image
Create a photo of a quiet countryside at sunrise, with rolling crossing an undulating river. An older man is walking his Irish red
green hills, a small cobblestone path, and an old wooden bridge setter in the foreground.
CHATGPT COPILOT
ChatGPT’s image Copilot initially
looked good on the presented its
surface, with all the image in a portrait
requested details, layout rather than
but despite the depth landscape, so we
of view it seemed a asked it to adjust
bit flat. Closer this to match the
inspection revealed other tools. More
the dog didn’t look like an Irish red setter and its back legs disappointing was the yellow sheen, which may have been to
appeared to be moving while its front legs were still. The paths reflect the glow of sunrise but made the result look unmistakably
in the background didn’t match the foreground and the river AI. The composition of the picture was very similar to
was flowing in two directions. ChatGPT’s, but with a more accurate Irish red setter.
GEMINI PERPLEXITY
Gemini’s image was Perplexity limits how
the only one to many images you
actually look like a can create for free
genuine photo, (we could only try
foregoing AI design once), and for the
flourishes to focus best results you need
on realism. We liked to pay for the
the buildings and the advanced models of
hills in the background, the detailed flora in the foreground and its Pro plan. It warned us “this question looks difficult” but still
the way the river opened out beyond the bridge. The man produced a visually appealing image. The man, bridge and
looked very lifelike, with a subtle smile and authentic clothes, landscape looked pretty realistic, but the dead-eyed dog and
and Gemini perfectly captured the panting Irish red setter. unnecessary mist were giveaway signs of AI.
Prompt 2: Combining two images
Combine these images so the subject of the first photo looks CHATGPT
trapped behind the gate of the second photo (both below). ChatGPT did a seamless
Replace the ‘private parking’ signs with ‘danger keep out’ and job of combining the
‘enter at own risk’. two images, but again
there was a flat quality
to the result that
stopped it from looking
real. This wasn’t helped
by how it positioned the
subject behind the
middle black bar, which
made it look like one
ear was separate from
his head. We were more
impressed by its
authentic-looking
signs.
OCTOBER 2026 APC MAGAZINE 39

