Can Opus 5.5 Build an Explainer Video Generator in One Prompt? I Tested It Three Ways

I know it's not perfect. I know it's not this crazy wow factor. But that's what Clearmud is all about: showing you the real results when you prompt from the perspective of a beginner.

In this one, I ran three experiments with Opus 5.5 on extra high. Every build was a single prompt. No fine-tuning behind the scenes, no polishing before I hit record.

• An explainer video generator, so you could start a faceless YouTube channel
• An app that turns IKEA assembly instructions into a visual lesson
• A blind test of five models inside the motion graphics tool my team built in house

Here's what each one gave me, and what I'd do next with them.

Build 1: The explainer video generator

This is the prompt I gave it, word for word. Feel free to copy it and rebuild it yourself.

> Build me an explainer video generator. I should be able to generate a script from an idea in chat form, then generate the video itself. It should be a blend of infographic, educational, visual learning, whiteboard, cartoon, motion graphics, and animated character and real stock media related to the script or generated images related to the script. Ideally, each video could be generated with a specific theme listed above or a blend of those themes. Please use our local Claude plus OpenAI subscriptions to power this end to end. It should also do sound effects and voice over with the option for me to record my own voice.

That prompt worked for 49 minutes, then came back and worked for another 12.

When I went to show it live, the app wasn't ready after a hard refresh. So I pulled up its default generations instead. Build in public means you show that part too.

### What it produced

The first video was a kids-style explainer on how bees make honey, with full voice-over. It opens with "One honeybee makes just 1/12th of a teaspoon of honey in its whole life," then walks through nectar, the honey stomach, the mouth-to-mouth passing at the hive, the wing-flapping to dry it out, and the wax seal.

The second video told the story of OpenAI from its December 2015 launch, naming the founders and the billion-dollar pledge from early backers. I stopped it early [03:39], but the point was clear. It mixed exactly what I asked for: a bit educational, a bit whiteboard, a bit cartoon.

Quick note: the facts in those two videos came from the AI, not from me. I didn't verify the honey math or the OpenAI history. That's part of what fine-tuning is for.

### Why this matters

This was one shot. Think about two or three more revisions. You'd essentially have a faceless YouTube video creator that you could start pumping out videos with once you take the time to fine-tune it.

And if you're doing this for your business or for advertisements, think about all the context and knowledge you have in your head. That's where you improve the initial prompt. Feed it everything you know. The output is only as specific as the brief.

Build 2: The IKEA visual builder

This one is such a simple idea, and it's one of my favorite things I've built lately.

The prompt:

> I have an idea. I want to create a visual product builder for IKEA stock and IKEA products. I want to turn the assembly instructions and documents that you would have to read page by page into a step-by-step visual lesson that allows the user or viewer to visualize it and see it without having to turn a page.

Then I told it to start with one product. It worked for 32 minutes.

### What it produced

It nailed it. There's a manual view and a realistic view so you can see the product itself. Then you hit "visualize the build" and it steps you through the whole thing [06:15]:

• Every tool you need, with the hardware counted off on the right
• Flip the panel
• Side panel holes, then the back slides in
• Insert the bottom
• Tip-over hazard warning
• Secure to the wall

One product, one prompt, 32 minutes. I didn't touch it after. This was a one shot, everybody.

Build 3: Five models, one four-word prompt

For the last test, I used a motion graphics tool my team built in house. We use it for clients and for our own content every day.

The entire prompt was four words: "how bees make honey." I ran it through five models and labeled the outputs A through E so I could rank them without knowing which was which.

### My blind ranking

• A: Too slow and too basic. Decent, but I didn't like it much.
• B: The bee flying, the flower evolving as it went. Really good. My favorite.
• C: Showed the wings drying out the top of the honey cell. My runner-up.
• D: Didn't wow me.
• E: A 15-second guide that rushed through everything. I wished I had more time to read it.

Final order before the reveal: B, C, E, then A and D at the bottom.

### The reveal

• B, my favorite, was Opus 5.5.
• C, second place, was Fable.
• E, third, was Sonnet. It rushed.
• A and D were the two GPT models. Not quite there yet for motion graphics.

I prompted four words and got that. There are no excuses anymore. You can jump to the reveal at [09:51] if you want to see the five side by side.

The takeaway: top dog changes weekly

At the time I recorded this, Claude is the top dog for this kind of work. In a week, it could be GPT. Then Gemini, then Grok. It's going to constantly shift, and I'm not going to pretend I know where it lands next month.

That's exactly why the model ranking is the least important part of this post.

What matters is that a beginner-level prompt now gets you a working starting point in under an hour. Not a finished product. A starting point. The gap between that and something you'd ship is two or three rounds of fine-tuning with your own context poured in.

And let me be honest about one more thing. You're not going to one-shot something that wows you. Nobody is. I'd estimate 80% of creators right now are fine-tuning by hand and then presenting the result as a one shot. I'm not here to do that. I'm here to show you what a prompt actually gives you.

What to try this week

• Copy the explainer prompt above. Swap in your own business context before you run it. Be specific about audience, tone, and what the videos are for.
• Pick one document people hate reading and ask for a step-by-step visual version of it, the way I did with the IKEA instructions. Start with one product, one process, one page.
• Run the same tiny prompt through more than one model before you commit. Rank them blind if you can. Your gut will surprise you.

So what are you building with AI right now? And what are you building from a motion graphics angle? Drop a comment on the video and let me know what you'd like to see me build next.

I'm not an AI expert. I'm building in public and sharing what actually works.

Watch the full video: https://www.youtube.com/watch?v=2rNgBw639do