ChatGPT Images 2.0 tested on branding, text and infographics: impressive, but not error-free

ChatGPT Images 2.0 is the biggest upgrade to image generation in ChatGPT in some time. A ZDNet journalist ran a series of hands-on tests and found that the model takes a big step forward, especially with text, branding and infographics, though not without errors.

What changed in Images 2.0

Text has been the main pain point of AI image generation for years. Earlier generations of models either avoided lettering in the frame entirely or produced an unreadable mush of pseudo-characters. ChatGPT Images 2.0 handles it in a fundamentally different way: headlines, chart captions and short blocks of text in infographics all come out readable and placed where they should be.

The second major shift is branding. The model understands requests to create visuals with a specific color code, set typography and brand elements. This opens up a new use case: rapid prototyping of marketing materials without a designer.

How the test went

The journalist tested Images 2.0 in turn on several practical tasks:

• Creating infographics with data, coordinate axes and labels
• Generating branded cards with text and a logo
• Reproducing a given visual style: colors, fonts, composition
• Preparing visuals for social media and presentations
• Text accuracy: headlines, bullets, numerical data

The overall verdict is "impressive, but sometimes wrong." Where the request is clear and the layout is simple, results are consistently good. But with non-standard fonts, complex multi-column grids or several text blocks at once, the model can still produce artifacts and shift elements to the wrong place.

Real-world use at work

Despite the errors, Images 2.0 is already usable for specific work tasks, which was not obvious even a few months ago.

Quick illustrations for content. For blogs, articles and internal documents, generation takes seconds instead of hours in Figma or Photoshop. The result is not final design, but it is a good enough draft for a first discussion.

Prototypes of marketing visuals. If you need to show a client or team the rough look of a banner, card or ad, Images 2.0 works as a first-draft tool.

Infographics without a designer. With the right prompt, the model structures the data and lays it out on the image. That used to require Canva, Piktochart or a separate task for a designer.

"It's more fun than I expected, and surprisingly useful for real

work," is ZDNet's takeaway from the testing.

Where problems remain

Images 2.0 works significantly better with specific, detailed instructions. Vague or overly general requests still often produce unsatisfactory results.

Another downside is unpredictability: the same prompt can give different results of varying quality across separate requests.

There is no direct control over parameters: you cannot rigidly define a grid, specify the exact placement of elements or use your own fonts. For professional design work, this remains a significant limitation.

What this means

ChatGPT Images 2.0 is not a replacement for designers and not a "Midjourney killer." But it is the first AI generator in a mainstream product that handles text and branding in images reliably and consistently. For anyone who wants to create a working visual quickly without special skills, it is already a functioning tool, not a demo effect.