Testing ChatGPT's new Image model (full review)

Описание: In this episode, I test-drive OpenAI’s brand new image model inside ChatGPT and share my first impressions. I walk through how to access it, how the “style” carousel works, and why the built-in prompt optimization matters if you care about output quality. I run a few practical experiments—plushie, graphite sketch, hand-drawn diagram, and a bobblehead to see how well it follows instructions and handles iterative feedback. By the end, I compare it to Nano Banana Pro (my recent daily driver) and give my honest verdict on whether it’s actually better

Timestamps:
00:00 – Intro
02:10 – Test 1: Plushie
04:10 – Test 2: Sketch
10:10 – Test 3: Bobblehead
12:30 – Final Thoughts and Conclusions

Key Points

I care less about how it works and more about whether it produces consistently great outputs.
The “style carousel” + automatic prompt optimization can be the real unlock for better generations.
The model’s ability to take feedback (clean edits without ruining the image) is the make-or-break test.
Turning complex ideas into clean “hand-drawn” diagrams is a practical use case for content that performs.
My takeaway: it beats my expectations, and it’s as good as (or better than) Nano Banana Pro.

Section Summaries

1. *What I’m Testing And Why It Matters*
I kick things off by explaining what I’m evaluating: whether OpenAI’s new image model can actually outperform Nano Banana Pro in real workflows, not just cherry-picked demos. I’m looking for output quality, reliability, and how well it takes feedback.

2. *How To Use The New Image Flow In ChatGPT*
I show the basic path: jump into ChatGPT, hit the plus button, click Image, and you’ll see a carousel of styles. That UI flow matters because it lowers the friction to experiment fast.

3. *Plushie Test + The Hidden “Prompt Optimizer” Layer*
I start with a plushie style test (using Sam Altman) and notice how ChatGPT effectively generates an optimized prompt behind the scenes. I connect it to tooling I’ve seen elsewhere (like Glyph App) and why prompt optimization can be as important as the model itself.

4. *Sketch Test: When Great Outputs Equal Real Leverage*
I run a sketch transformation using a photo of me drinking a martini and zoom in on the details. I’m blunt about the real reason I care: better outputs can mean better ads, better content, and a higher chance my businesses win.

5. *The Feedback Test: Can It Edit Without Falling Apart?*
I push the model with revisions—like removing elements that make it feel “too AI” (the hand/notebook vibe). This is where a lot of image models fail, and it’s why I’ve liked Nano Banana Pro: it tends to take iterative direction better.

6. *Hand-Drawn Diagram Mode: The “Viral Diagram” Use Case*
I try converting a diagram I posted on X into a more natural hand-drawn style and explain why this matters: these simple visuals can communicate complex ideas and tend to perform. The output surprises me—in a good way.

7. *Bobblehead Test + Productization Thoughts*
I test a bobblehead variant and add constraints (no baseball uniform; make it feel like a tech YouTuber). Then I zoom out to the next logical step: if you can generate designs like this, how do you turn them into real products—manufacturing and a Shopify store included.

8. *Blog Post Claims + My Final Verdict*
I reference OpenAI’s blog claims around editing (add/subtract/blend), creative transformations, instruction following (like the grid example), and better text rendering. My conclusion is simple: it beat my expectations, and I think it’s as good as—if not better than—Nano Banana Pro.

The #1 tool to find startup ideas/trends - https://www.ideabrowser.com

LCA helps Fortune 500s and fast-growing startups build their future - from Warner Music to Fortnite to Dropbox. We turn 'what if' into reality with AI, apps, and next-gen products https://latecheckout.agency/

The Vibe Marketer - Resources for people into vibe marketing/marketing with AI: https://www.thevibemarketer.com/

FIND ME ON SOCIAL

X/Twitter:   / gregisenberg  
Instagram:   / gregisenberg  
LinkedIn:   / gisenberg