Image tools paint letters, so words come out wrong. Code types them, so they can't. Here's when to use ChatGPT, when to use Claude, and the exact asks for both.
You make a graphic for a post. It looks beautiful, and one word is spelled wrong. You regenerate, and now a different word is wrong. That happens because image tools paint pictures of letters, like an artist painting a sign, and painters slip.
Claude takes a different road. Ask it for an infographic and it writes a tiny web page: real typed text, real numbers, laid out in code. You screenshot the result. Typed text can't be misspelled, and when a number changes you edit one line instead of re-rolling the whole image.
Mood, texture, and visual richness are what image models do best. Use it when the picture matters more than the data and the text on it is minimal.
Prices, stats, tip lists, process steps. Code never misspells and never rounds your numbers creatively. This is the teaching-content workhorse.
ChatGPT paints an artistic background and leaves empty space. Claude puts the real text and charts on top. The premium look with none of the spelling gamble.
For any graphic carrying data, a list, or steps. Use the regular chat, and upload 1 to 3 reference images of a style you love first if you have them.
Why it works. Naming HTML tells Claude to build, not paint. The screenshot instruction gets you a clean image without touching code yourself.
Build me an infographic as a single HTML page, sized [1080x1080 for a feed post, or 1080x1350 for portrait]. The content: [your headline, your list or numbers, your name or handle]. Match the style of the reference images I uploaded. Every word and number must be typed text, spelled exactly as I wrote it. When it's done, show me a preview and tell me how to save it as an image.
For the artistic piece, or the background half of the hybrid. The empty-space line is the part people miss.
Why it works. Asking for empty space reserves room for the real text layer, so the background never fights your words.
Create an image: [describe the scene, style, and colors you want], in [square / portrait] format. Leave generous empty space at [the top / the center] for text and charts I'll add later. No words or letters anywhere in the image.
Download the ChatGPT background, then bring it to Claude with this ask.
I've uploaded a background image. Build an HTML infographic that uses it as the full background and lays this content on top in the empty space: [your headline, numbers, list]. Typed text only, high contrast so everything is readable, nothing covering the important parts of the image.
When you want the same design on a new topic, press Edit on your FIRST message in the chat, swap the topic, and send. The whole design rebuilds around the new content, same style, no redesign.
Once Claude nails a look you love, you can stop describing it forever. Have Claude bake the style into a skill, and future graphics take one line.
I love the style of the infographic we just made. Create a skill called [my-graphics] that saves this exact design system: the colors as hex values, the fonts, the spacing, the layout logic. When I use the skill with new content, build the new infographic in this same style without me describing it again. Walk me through saving the skill to my account, step by step, beginner level.
A style skill buys consistency and costs variety. Perfect for a recurring series that should look like family. For one-off pieces, skip the skill and let each graphic match its content. If your feed starts looking repetitive, that's the signal to add a second style skill, not to force the first one everywhere.