Claude Design arrived and instantly took the world by storm. Phrase on the road is that UI designers are already altering their LinkedIn titles to “immediate engineer.” I’m joking (principally).
However Claude Design is genuinely spectacular. Across the identical time, OpenAI upgraded its already frontier-level picture technology in ChatGPT with ChatGPT Pictures 2. It could possibly comply with prompts with significantly better accuracy, and it appears to have mounted one of many greatest issues with AI photographs: textual content.
NotebookLM, in the meantime, has been within the text-heavy picture technology sport for some time. It could possibly generate slide decks, infographics, and different visible summaries out of your supply materials.
Textual content has all the time been a wierd weak point for picture fashions. They generate textual content inside a picture, which is already a bit problematic, as a result of textual content is one thing you’re supposed to jot down. Claude Design isn’t precisely in the identical class as ChatGPT Pictures or NotebookLM, however I nonetheless needed to place it to the check. They are saying it may well design something, in any case.
I needed one thing visible, however textual content heavy. And what higher check than an infographic?
Associated
I requested Claude, ChatGPT, and Gemini to construct a simulation, and one winner was apparent
The LLM race stopped being an in depth contest fairly rapidly.
Organising the showdown
We wish a cool infographic
Amir Bohlooli / MUO
The check is straightforward: design an infographic for the Raspberry Pi 4 Mannequin B.
It’s an excellent benchmark. A blueprint-style tech infographic is dense with particulars. There’s a whole lot of textual content, a whole lot of labeling, and a whole lot of alternatives to get issues subtly improper. That makes it excellent for seeing which software can write textual content correctly, which one can level that textual content on the proper shapes, and which one can really draw these shapes appropriately within the first place.
NotebookLM’s infographic software in Studio makes use of Gemini’s picture technology. ChatGPT makes use of ChatGPT Pictures 2. Each of these are picture fashions.
Claude Design is completely different. It makes use of Opus 4.7, which is a textual content mannequin, not a picture mannequin. In different phrases, Claude Design is actually simply the identical Opus 4.7 you’ll be able to chat with, besides it’s been given instruments and fine-tuned for UI design. Claude Design is the agent; Opus 4.7 is the mannequin. So as an alternative of producing a picture within the standard sense, Claude Design builds the requested visible in HTML and CSS.
I needed to check it anyway. Possibly Claude Design has some tips up its sleeve. Possibly it may well generate or assemble its personal property effectively sufficient to compete. There was just one solution to discover out.
For the check, I gave all three instruments the identical supply materials: a GPIO diagram of the Raspberry Pi 4, a textual content file with the Pi’s technical specs, and the official Raspberry Pi product transient PDF. Then I gave all of them the very same immediate:
Create a blueprint-style infographic for the Raspberry Pi 4 Mannequin B utilizing a darkish navy background, white and cyan line artwork, and a delicate grid overlay. Embody a quick introduction overlaying what it’s and who it is for, a key specs block, an annotated top-down board diagram with labeled parts, a color-coded 40-pin GPIO pinout, and a use-cases part. Use leader-line callouts, dashed part dividers, and a monospaced technical font all through, specified by A3 panorama orientation.
After which we choose the outcomes. Let’s see what occurred.
NotebookLM
Fairly, however very improper
Amir Bohlooli
NotebookLM has been making slides and infographics lengthy earlier than the opposite two, so now we have pretty excessive expectations. I may have requested it to analysis the Raspberry Pi 4 by itself and cite its sources, however for a fairer check, I created a brand new pocket book and gave it the identical three recordsdata as everybody else.
NotebookLM completed first, and at first look, the outcome was visually spectacular. It regarded cool. The blueprint type was there and it included many of the sections I requested for. However the nearer I regarded, the more severe it acquired.
The generated board schematic is kind of recognizable, however the labeling may be very improper. I’ve walked by means of the problems within the gallery above, however the brief model is grim. The 2 micro-HDMI ports are drawn appropriately, however certainly one of them is labeled as a USB 3.0 port. Beneath them, the audio jack is labeled as a USB 2.0 port. In the meantime, the precise USB ports subsequent to the Ethernet port are labeled because the digicam and show ports. Tragic.
Amir Bohlooli
NotebookLM’s design additionally features a random price-to-performance comparability that doesn’t actually make sense, however that’s nothing in comparison with the larger downside.
The true dealbreaker is the GPIO pinout. On the diagram, NotebookLM labels it as a “48-pin” header, then within the pinout part, it calls it “40-pin.” That might already be dangerous sufficient, however the pinout itself is worse. Many of the labels are improper, and the remaining are mangled gibberish. Look carefully on the numerals contained in the pins and also you’ll see the same old AI-image melting.
NotebookLM made a powerful visible. However it didn’t make a usable infographic. You couldn’t print this out and depend on it. You’d fry your board.
Claude Design
Sensible labels, dangerous board
Amir Bohlooli
Claude took the longest to complete. When it was finished, the outcome regarded spectacular from a distance, however the phantasm broke sooner than NotebookLM’s. The board merely didn’t appear to be a Raspberry Pi 4.
That solutions our query: Claude Design can’t actually generate photographs. It could possibly design round photographs, prepare property, and construct layouts, however when requested to create a technical board diagram from scratch, it attracts one in code. And on this case, that drawing was a “Raspberry Pi-inspired rectangle” than a Raspberry Pi 4.
The structure had issues, too. The orientation was improper. On the actual Raspberry Pi 4, the USB stack sits throughout from the microSD slot. In Claude’s model, the GPIO header is there as an alternative.
However when you look previous the board form, Claude does significantly better than anticipated. The ports are appropriately labeled, and the callouts embrace helpful, correct particulars. It clearly understood what the parts have been speculated to be, even when it couldn’t draw the board itself correctly.
Probably the most spectacular half is the GPIO pinout. It’s dead-on correct. The numbering is right, the labels are right, and the colour coding is correct, too. It additionally appears clear and readable. Applause to Claude!
To this point, we would have an ideal infograph if we took out NotebookLM’s picture of the board and fused it with Claude’s work. I needed to attempt that, however I used to be afraid I might run out of Claude Design credit.
Claude was so happy with its work that it even signed the design.
ChatGPT
We’ve got a winner
Time for the final check. Like NotebookLM, ChatGPT generates its outcome as a picture, this time utilizing ChatGPT Pictures 2. It took about 5 minutes to complete, which put it neatly within the center: slower than NotebookLM, sooner than Claude Design.
At first look, the outcome follows the identical basic shade palette and design language because the others, however with one large distinction: the board isn’t drawn like a blueprint. It’s rendered as an precise Raspberry Pi 4 board. That’s fantastic, nevertheless it additionally means ChatGPT made the duty more durable for itself, as a result of it’s a lot simpler to mess up a sensible board than a simplified diagram.
We have got all of the stuff we requested for, intro, tech specs, use circumstances, and the pinout. I believed I might be up for disappointment once I zoomed, however, I used to be shocked as an alternative. The picture is actually correct.
The HDMI ports are labeled appropriately. The audio jack is labeled appropriately. The USB ports are labeled appropriately, too. The one actual mistake is that the USB ports closest to the Ethernet port ought to be the USB 3.0 ones, however ChatGPT assigns USB 2.0 there as an alternative. Additionally, each ports are blue inside. I’m nitpicking at this level.
The extent of element is spectacular. The Ethernet port (which ChatGPT calls Ethernet jack. Who does that?) has an imprint similar to the actual factor. We have even acquired the little Made within the UK print. Neat!
The one labeling downside with this design is the CSI digicam connector, which is pointing at nothing.
Past that, the GPIO pinout is superb. The numbering is right, the labels are right, and the features are right. Even the colour coding is correct! That’s significantly spectacular. The one barely gooey half is the little header within the center, I am unsure what that is speculated to be.
ChatGPT wins. You could possibly really use this infographic (in the event you delete the digicam connector label).
Proper software for the job
For this job, ChatGPT is the best software. The primary picture improve in ChatGPT already blew us away, however ChatGPT Pictures 2 does an excellent higher job with small particulars, and it hallucinates a lot lower than the opposite picture fashions I examined. Despite the fact that the infographic was generated as a picture fairly than written as a doc, I didn’t spot a single goopy, gibberish letter pretending to be textual content.
NotebookLM was the frustration right here. You’d count on a picture software inside NotebookLM to be particularly good with textual content, even when the visuals have been a bit weaker. Someway, I acquired the other. It drew a handsome picture, however tousled the labels, which is pretty deadly when your entire level of an infographic is to tell.
Claude Design is a distinct case. It’s clearly not the best software for this particular job, however as an experiment, it was nonetheless helpful. Its greatest benefit is one thing neither of the opposite instruments provide: it designs in code and provides you an open-layer output. Which means you’ll be able to obtain the HTML and edit it your self, have one other AI revise it, and even take the higher board picture from NotebookLM and use Claude Design to construct a cleaner structure round it.
Sadly, Claude Design can be so costly that it’s borderline unusable except you realize precisely what you need earlier than you immediate it. I’m on the Claude Professional plan, and the only immediate I used above ate up 25% of my weekly credit. That makes the “proper software for the job” query matter much more. Use ChatGPT to discover concepts, generate visible instructions, and check ideas. Then, after getting one thing stable, usher in Claude Design to assemble it into one thing editable.
Developer
Anthropic PBC
Value mannequin
Free, subscription obtainable
Claude is a sophisticated synthetic intelligence assistant developed by Anthropic. Constructed on Constitutional AI ideas, it excels at advanced reasoning, subtle writing, and professional-grade coding help.
OS
Android, iOS, Net
Developer
OpenAI
Value mannequin
Free with non-compulsory subscription
ChatGPT is the flagship AI chatbot from OpenAI, and it is loaded with options.
OS
Android, iOS, Net-based app
Developer
Google
Pricing mannequin
Free
NotebookLM is Google’s AI-powered analysis pocket book that reads what you add and helps you remodel it into structured summaries, explanations, and visuals.
