Fake a ChatGPT thread without opening ChatGPT
Say you’re writing a tutorial on prompt engineering and you need a clean shot of the model explaining recursion. Screenshotting the real app means logging in, waiting on a reply you can’t fully steer, then cropping the sidebar out. Skip all that. Type both halves of the conversation here and the canvas paints a ChatGPT thread that reads like the genuine article.
Each message gets drawn pixel by pixel with the HTML5 Canvas API. User turns land in the gray rounded bubble on the right. Assistant turns sit on the left as plain text next to the little green OpenAI mark, exactly how the web app stacks them. No screen recording, no browser extension, no html2canvas trickery.
What you get to change
The whole conversation is yours. Add a turn, delete one, shuffle the order with the up and down arrows. Flip any message between “User” and “ChatGPT” from the dropdown.
That pill up top is the model switcher. Type “ChatGPT 4o”, “ChatGPT 5”, “GPT-4.1”, whatever suits the shot, and it renders bold with a little chevron beside it. Don’t want a header at all? Uncheck it for a cropped, message-only frame.
Dark mode is the charcoal #212121 people recognize on sight. Light mode is clean white, kinder to print and slide decks. One checkbox flips everything at once: background, text, bubble shade.
A few things worth knowing
Long replies wrap themselves. Paste a three-paragraph answer with line breaks and the canvas measures every line before it draws, so the image grows to fit. Nothing clips off the bottom.
The export is a PNG at twice the display resolution, roughly 1200 pixels wide. Sharp on a retina screen, sharp in a zoomed presentation, sharp on a printed handout. You get the full-res file, not a fuzzy preview grab.
Here’s the privacy part: none of your text touches a server. The rendering runs on your own machine, so a draft conversation about an unreleased feature stays on your machine.
Who’s using it
Teachers building slides about how language models respond. Marketers who want an AI exchange for a landing page without wiring up a live key. YouTubers making thumbnails. Bloggers illustrating an article on AI tooling. And yeah, meme makers, that’s honestly half the traffic.
One ask: don’t pass a fabricated screenshot off as a real statement from OpenAI or anyone else. Mockups are for illustration, keep them labeled that way.
FAQ
Does this actually talk to ChatGPT?
Nope. You write both sides yourself. There’s no model behind it and no API call, which is the point, total control over every word.
Is my conversation private?
Yep. The image is drawn locally with the Canvas API. Nothing you type leaves the browser tab.
Can I change the model name in the header?
That field is fully editable. Set “ChatGPT 4o”, any custom name, or hide the header entirely for a bare crop.
How many messages can I add?
As many as you want. The canvas resizes to fit the full thread, two turns or twenty.
What size is the download?
It exports at 2x, around 1200px wide, with the height set automatically by your content. Stays crisp when zoomed.
Does it do dark mode?
Yes. Flip it to match ChatGPT’s dark charcoal, or leave it light for a print-friendly version.