What You’ll Learn

Mythos 5 (and Fable 5, its “on a leash” sibling) landed, and for anyone creating educational content, the change in how AI generates visuals is dramatic. In this session James runs a head-to-head: he takes the same text-heavy lesson and asks Sonnet to build a visual asset, then asks Mythos — and the results are night and day. This walkthrough shows the comparison and what it means for reviving your old courses.

By the end you’ll understand why the new models change the game for course visuals, and how a single prompt can turn a wall of bullet points into an interactive learning asset embedded right in the lesson.

The Problem: Your “Visual” Lesson Is Really Just Text

James pulls up a real lesson from his campus — one built about seven months ago to help people choose a NotebookLM infographic style. The catch: it describes ten visual styles entirely in text and bullet points. There are no actual images of the styles, so a lesson about visual languages contains zero visuals.

“Here’s an opportunity where we have a whole bunch of text that is representing something visual.”

💡 In Plain English: it’s like teaching color theory with a page of words. The information is right, but the format fights the lesson.

The Test, Round 1: Sonnet

James gives Sonnet a single prompt: pull this lesson, look at its content, and turn it into an interactive learning asset because students keep getting stuck here. Two problems show up. First, even given the URL, Sonnet grabbed the wrong lesson. Second, what it produced was an HTML page that’s really just text — colors and bullet points, nothing you’d actually learn a visual concept from. Fine as a starting draft for a web page, but not a teaching asset.

The Test, Round 2: Mythos 5

Same task, handed to Mythos. The difference starts with comprehension. Mythos read the lesson and reported back: this is a lesson about visual languages that contains zero visuals, and it’s structurally flat — 30 near-identical lists. In other words, it diagnosed exactly why the lesson was weak.

Then it built the right fix on its own. Rather than dumping ten screenshots into the lesson, it created an interactive visual style picker — a gallery where each of the ten styles is rendered in its own style, and the learner answers two questions (what are you explaining? what feel do you want?) to find the best match. Pick “a system or platform,” and it highlights the Minecraft-style block with its rationale; pick “step-by-step process,” and it surfaces an animated machine-and-input graphic. Options that don’t fit simply don’t light up.

All of it was a single shot — one prompt, done, based on the lesson’s context — and Mythos even uploaded the standalone interactive HTML asset directly into the lesson in the course.

Why This Matters for Educators

The bigger realization: this isn’t limited to courses. You can point Mythos at a brief, a PDF, a slide deck, a talking-head video, audio, or a Zoom recording and ask it to generate interactive visual assets from the content. For anyone sitting on years of material — James mentions 80-plus courses — it’s a way to revive dead lessons by having the model read them and build the visuals they always needed.

“Which one do you want? Something like this, or something like this? That’s the change that’s here.”

The Takeaway

The old pain of making good visuals — hunting for a template in PowerPoint or Google Slides, or wrestling with Canva — is fading. The new models read the context of a lesson, decide the best way to represent it, and build an interactive asset in one prompt. Start testing it on a lesson that’s always been too text-heavy. If you want the whole system that runs James’s business around this, the campus operating system is at trainingsites.io/os.

Teach more, and let the agents do the rest.

This tutorial is a recap of a live demo.