OpenAI’s New Image Generator Is Very Good at Making Very Convincing Nonsense

You’re scrolling LinkedIn on a Tuesday morning. A post catches your eye — a crisp, professional infographic breaking down Q3 market trends, complete with clean bar charts, a polished headshot of the author, and a pull quote in a tasteful sans-serif font. You almost keep scrolling. Then something feels slightly off. The numbers don’t quite add up. The author’s eyes are a little too symmetrical. The chart’s Y-axis starts at a suspiciously convenient number. You’ve been had. Again.

This is the world OpenAI is accelerating into with ChatGPT Images 2.0, announced on Tuesday. And as someone who spends most of his working hours testing AI tools so you don’t have to, I have some thoughts — not all of them flattering.

What Actually Changed

Let’s be straight about what OpenAI shipped. ChatGPT Images 2.0 is a meaningfully better image generator than what came before it. The outputs are more realistic, more detailed, and — this is the part that matters — less obviously AI-generated than previous versions. Users can now generate multiple images from a single prompt, and the model has improved ability to follow complex instructions, including producing charts and data visualizations.

That last part is genuinely useful for certain workflows. If you’re building a presentation deck, mocking up a report, or need a quick visual for a blog post (hi), faster and more accurate image generation from a single prompt saves real time. I won’t pretend otherwise.

The model is rolling out through OpenAI’s flagship ChatGPT interface and also through Codex, its AI coding assistant. That’s a wider distribution than previous image tools, which means more people will have access to it faster.

The Part Nobody Wants to Say Out Loud

Here’s what the press release doesn’t address: making AI images harder to detect as AI images is not a neutral technical achievement. It’s a choice with consequences.

The original wave of AI-generated content was easy to spot. Melted hands. Six fingers. Text that looked like it was designed by someone who had only heard of the alphabet. That weirdness was annoying, but it also functioned as a natural filter. People learned to clock it. The tell-tale signs became cultural shorthand.

ChatGPT Images 2.0 is specifically designed to sand those tells down. OpenAI’s own framing — “smarter and more precise” — is doing a lot of work in that sentence. More precise at what, exactly? At producing images that slide past your skepticism before you’ve had your second coffee.

Some critics online have already labeled this an acceleration of “AI slop” — a term for the flood of low-effort, AI-generated content that clogs feeds, inflates content farms, and erodes trust in visual media. The counterargument, usually from people who sound defensive about it, is that critics are just tech tabloid writers who can’t help themselves. Maybe. But dismissing the concern doesn’t make it go away.

Who This Actually Helps

I review tools. My job is to tell you what works and what doesn’t, not to moralize at you. So here’s the honest breakdown.

• Solo creators and small teams who need quick visual assets will find real value here. Faster generation, better quality, multiple outputs from one prompt — that’s a solid workflow improvement.
• Marketers who need to prototype visuals before commissioning real design work have a genuinely useful tool.
• Developers building products on top of OpenAI’s API get a more capable image layer to work with.

Where it gets murkier is everywhere else. The same capabilities that help a solo blogger mock up a header image also help a content farm produce thousands of fake product reviews with convincing lifestyle photography. The same chart-generation improvements that help a legitimate analyst also help a bad actor fabricate data visualizations that look credible at a glance.

My Actual Take

ChatGPT Images 2.0 is a technically impressive update. The quality jump is real, the multi-image generation is useful, and the improved instruction-following makes it more practical for professional use cases. If you’re already in the OpenAI ecosystem, it’s worth trying.

But I’d push back on the framing that better AI image generation is straightforwardly good news. The harder these images are to detect, the more work falls on readers, platforms, and journalists to compensate. That’s a cost that doesn’t show up in the product announcement.

OpenAI built a more capable tool. What the space does with it is a separate question — and one that a Tuesday press release isn’t going to answer.

🕒 Published: April 23, 2026