I Put Claude's New Visual Tools to the Test Against ChatGPT

Claude has always been focused on text and code, but a few days ago, it rolled out image-generation capabilities—ways to get graphics, mock-ups, slideshows and other similar materials created by AI. The suite of tools is called Claude Design.

Then ChatGPT, already much more invested in AI image generation, rolled out a substantial update called Images 2.0. Promises were made of a "step change" in accuracy, consistency, and instruction following.

So what can you now do with these AI tools that you couldn't before? And how do the capabilities compare?

Claude Design is focused on business and enterprise

Claude Design is available for Claude subscribers, and is intended, in the words of Anthropic, for "visual work like designs, prototypes, slides, one-pagers, and more." It won't produce pictures of cats riding skateboards for you, but it will pull together a project slideshow, or a mock-up of an iPhone app.

This continues Claude's focus on business and enterprise: As well as coding with Claude, teams can work up prototypes and put together pitch decks. That may sound a little dry, but the new capabilities are actually quite versatile, and include spinning, interactive globes, as seen in the demo video.

To get started, you can get Claude to create visuals from an existing codebase, load up existing images and documents to use as starting points, or just type out a text prompt. You need to head to a special Claude Design landing page, separate from the main chatbot interface, which lets you choose how you want your workflow to operate.

I decided to build a slide deck showing off the value of Lifehacker, and gave the AI a few screenshots to work with for an idea of the style. Claude then asked me a few questions about what I wanted, including the mix of text and images, and how long the slideshow should be, before getting to work. You get to see the AI "thinking" and working through the steps of building the graphics in real time.

When the finished work was presented, it was impressively polished—and Claude Design gives you everything you need to export your work somewhere else. One of the most useful features is the way you can tweak the visuals after they've been created—on my slide deck, I was able to tweak the accent color, fonts, and slide density with just a few clicks.

You can also request edits via further prompts, and even draw on the visuals to indicate what should happen next. Moving on to an iPhone mock-up of a potential Lifehacker news app, Claude Design did a great job here too, bar one or two little graphical glitches: the app design that looked very Lifehacker-y, and I was able to request edits just by drawing on the visuals and typing out what I wanted to change.

It's all slick, professional, and easy to get around; I could see a lot of companies using Claude Design right alongside Claude Code. For individuals, it looks like a useful way of putting together ideas for designing just about everything, including slide decks (though the AI tools inside apps like PowerPoint and Google Slides might suit you better).

ChatGPT Images 2.0 is focused on consumer as well as business use

As for ChatGPT and its Images 2.0 upgrade, it's much more generalized and focused on consumer as well as business use. OpenAI says prompt instructions are now more closely followed, end results are more accurate and consistent, and text rendering has been further improved. Tasks can be more complex, and images look more "intentionally designed" as well.

Creating images is as straightforward as it always has been: Just click Create an image and explain what you want to see in the prompt box. People have been making complex Where's Waldo? images, infographics from scientific papers, and mock magazine covers; I was able to produce a quick comic strip about Lifehacker in minutes.

ChatGPT also proved able to mock up two Lifehacker magazine covers, of varying quality: They certainly look realistic enough, and there are no obvious mistakes or typos, but at the same time they also have that generic feel that a lot of AI imagery comes with. You can tell that these covers represent the "averaging out" of all the magazine covers sucked up in ChatGPT's training data.

You can't build slide decks or anything as complex as you can in Claude Design with ChatGPT Images 2.0. You can theoretically create single slides and app mock-ups, but there are many more limitations in terms of consistency and editing what's on screen afterwards—ChatGPT is much more about one-off moments of AI art.

The new Images 2.0 model is also much better at pulling up real information from the web, so you can put together a cartoon map of Middle Earth (although it's copyright-savvy enough to not do an exact copy of Tolkien's work), or create an informative diagram about upcoming sports tournaments. I ran both those tests, and ChatGPT came up with impressive visual results.

Rather than ChatGPT, the closest comparison to Claude Design is probably the Gemini AI tools available through Google Docs, Sheets, and Slides, which I've written about previously. While you can't yet create entire presentations, you can build slides from simple prompts, and load in other materials as references for the design.

As with Claude Design, you can use follow-up prompts to refine certain aspects of the slideshow or document design, without having to start again from the beginning. And your finished work can be exported into a variety of formats, including PDF and Microsoft Office-compatible file types.