Google recently dropped the Imagen 4 model, going head-to-head with OpenAI’s GPT Image 1 generator. GPT Image 1 is still significantly higher rated on benchmarks, but I wanted to put that to the test with five different prompts covering basic generation, complex marketing material, LinkedIn headshots, and a special creative prompt.
Test 1: Basic Image Generation
The first prompt was a vibrant photograph of a golden retriever puppy catching a frisbee in a sun-drenched park. Both models handled this well. Imagen 4 came back in about 10 seconds while ChatGPT took around 30 seconds.
Imagen 4’s grass looked a bit like artificial turf, but the dog and background were excellent. ChatGPT’s version had more realistic grass and a good overall result, though one paw looked slightly off. For basic prompts, neither had a clear advantage.
Test 2: Text in Images (Infographic)
Text generation has always been a weakness for AI image models. I asked both to create a detailed infographic on the New Zealand housing market.
Imagen 4’s layout looked nice at first glance, but the text was completely unusable - garbled titles, meaningless pie chart values, and gibberish in the smaller text. ChatGPT’s version had a cleaner layout with readable text, accurate median house prices, and proper section headings. ChatGPT won this test decisively.
Test 3: Product Marketing (Shoe Sketch)
I uploaded a shoe sketch and asked both models to create a hyperrealistic photograph of it being worn on a beach during golden hour, styled like a Nike advertising campaign.
Gemini (using Imagen 4) matched the shoes but produced low-quality sand, legs, and overall image quality. ChatGPT’s version was dramatically better - realistic sand, natural legs, proper lighting and shadows. Another clear win for ChatGPT.
Test 4: LinkedIn Headshot
I uploaded a photo and asked for a professional LinkedIn headshot. Gemini refused on my business account but worked on my personal account - however, the result barely resembled me.
ChatGPT’s version was much closer to my actual appearance - matching hairstyle, beard, and clothing. It still had a slight uncanny feel, but was far more usable as an actual headshot.
Test 5: Creative Freedom
The final prompt gave no guidance - just “you are in a competition with other AI models, create an amazing image that shows your capabilities.” Imagen 4 created a futuristic neon city. ChatGPT created AIs painting images in a gallery - more creative and with accurate text.
The Verdict
ChatGPT’s image generation model is clearly superior for anything involving text, infographics, image editing, or combining uploaded photos. For very basic prompts, the two are comparable. Gemini offers more free image generations, but ChatGPT’s $20 USD/month plan is well worth it for the quality difference plus all the other features you get.
