Image Test - Grok vs Gemini vs ChatGPT

Let's do some image generation tests with some of the leading AIs

Put the leading AIs to the test with their Image generation capabilities. To be fair, here are the rules:

  • I will use their basic non-paid versions only

  • I use the standard processing mode, no advanced modes

  • Exact same prompt each time, no variations.

  • I don't use the prompt suggestions made by the AI. I stick to this rigidly.

  • I don't allow the AI to rely on memory or retained knowledge of anything. All are create with a blank slate.

Let's start with creating a simple image:

"Produce an image of a beautiful brunette woman, full length, photorealistic, she is wearing a green and orange sequin dress, she has a 1970s vibe. Mobile Phone 9:16 format"

Drawing people

Gemini

ChatGPT

Grok AI

Very interesting comparison. I was perhaps intentional in not including a lighting or background preference and just allowed the AIs to make this up on its own.

The detail and colours in the Gemini stand out a bit more than the others, which shows it leaned into the "photorealistic" aspect and created a virtual environment or setting for the character created.

Grok went for a a more sultry option with a plain background, preferring to spend more time on making the skin tones a realistic as possible, and ensuring that the character fit the "1970s" aesthetic.

ChatGPT's image looks like a magazine photoshoot, and perhaps draws on these as a template for creating this image. It's a bit darker in terms of brightness than I would like, but it might be setting a mood of sorts.
Now let's notch up the difficulty a little bit. This time we are asking the AI to draw a city scape, but also to include wording in the middle. The test is both the scene and also getting both a haiku right and also ensure the wording is correct.

"Draw an image of a busy city at street level. It is night time. There is a neon sign in the middle with a short haiku with an ironic message. Mobile phone 9:16 format."

Drawing a city with words

Gemini

ChatGPT

Grok AI

Pretty surprised that all three AIs decided that "neon sign" means to locate the city in the East Asian region. Perhaps this is all it could find as a template? Or perhaps western countries don't do neon signs? I was surprised we didn't get any Vegas-esque cityscapes which would have been the obvious choice.

The Gemini image again created an image with the most bright and vibrant colours. Perhaps ignoring the "night time" detail. The haiku isn't bad but the sign sort of floats and doesn't look attached to anything.

The ChatGPT one is again pretty dark and details are hard to see. It did a pretty good job on the haiku too with no spelling errors. Seems to spam the word "Regal" a lot in the background which is disappointing.

Grok AI did a very photorealistic image which I didn't ask for, but do appreciate. No idea as to the haiku and if it makes sense. I guess given no direction on the haiku language it defaulted to the Japanese language.
Let's get surreal by asking for a realistic photo of something abstract. Adding armour to the design will try to steer the AI from lazily drawing from existing drawings and just copying them.

"Draw a photorealistic image of an armoured Jabberwocky. Mobile phone 9:16 format."

Drawing abstract

Gemini

ChatGPT

Grok AI

Quite similar fantasy settings here, much removed from the traditional Alice in Wonderland design except they all have the same dragon-type aesthetic.

The Gemini image again created an image of a humanoid dragon. Might have leaned more into the "armour" aspect than using the traditional references. Looks pretty good and improvised the axe and fantasy background.

The ChatGPT one captures what I imaged pretty perfectly as I wanted a Jabberwocky wearing dragon-type armour instead of an armoured knight with a dragon head.

Grok AI did something a little unexpected. Perhaps it researched the traditional description and then get confused by the amour references.