In this article
What happens when you give AI a few instructions, some existing design assets, and let it create an entire promotional video? We decided to find out.
AI Can Make Videos Now?
The recent Claude Opus 5.5 update has been pretty impressive.
We've been using AI for coding for quite some time, but its capabilities are expanding beyond that. From creating 3D scenes in Blender to producing animated videos, Claude is becoming an increasingly interesting tool for creative work.
Naturally, we wanted to try it ourselves.
Instead of experimenting with random concepts, we decided to use two of our actual projects at OSBR to see how far we could push AI-powered video creation.
Experiment 1: Creating an OSBR Promotional Video with Just Claude
For our first experiment, we wanted to create a short promotional video for our company website.
We kept things simple: just Claude Opus 5.5, without any additional frameworks or plugins.
Here's the interesting part: we didn't provide any screenshots, screen recordings, or visual references.
We simply asked Claude to use the existing assets from our website repository, follow our branding, and create a video of a specified duration showcasing our company values and what we do.
Three Prompts, 20 Minutes
The first result was... okay.
There were some overlapping visual elements, the pacing felt a little strange, and certain transitions needed improvement.
For the second prompt, we simply pointed out the overlapping elements and asked Claude to fix them. Surprisingly, the result was noticeably better.
By the third prompt, we were mostly tidying up the pacing and making small adjustments.
And that was it.
Three prompts. Roughly 20 minutes. One pretty decent promotional video.
Unfortunately, we couldn't find the earlier versions for a before-and-after comparison, but here's the final result.
It's not perfect, but considering how little time and effort went into it, we're pretty impressed.
Experiment 2: Taking Things Further with Hyperframes and Bungkus-mc
For our second experiment, we decided to introduce Hyperframes, a plugin that provides additional tools for AI-powered video creation.
We've been actively updating bungkus-cli, our in-house project scaffolding tool, and recently created a new mascot for our latest product, bungkus-mc.
It seemed like the perfect opportunity to create a promotional video.
We followed the same approach as before. No screenshots, no recordings, and no detailed storyboard. Just our existing Figma assets, which included a character we'd designed manually, and a few instructions about the features we wanted to showcase.
The Difference with Hyperframes
The result genuinely surprised us.
That's not to say Claude on its own wasn't good, but Hyperframes opened up more possibilities for animations and video production.
One feature we particularly liked was its local editor, which allowed us to make manual adjustments instead of relying entirely on prompts.
Sometimes, it's simply easier to move an element yourself than explain the same adjustment to AI repeatedly.
I can't remember the exact number of prompts it took to achieve this result, but it was definitely fewer than five.
Not bad for a video created almost entirely through prompting!
Final Thoughts
The improvements we've seen with Claude Opus 5.5 are genuinely exciting.
While Fable 5 remains our preferred model for certain complex tasks, Opus 5.5 is becoming a strong alternative. In our experience, it delivers impressive results across coding and creative tasks, including 3D modelling and video creation, while potentially using fewer tokens depending on the workflow.
What excites us most is that AI is no longer just a coding assistant. It's becoming a tool that can support the entire creative process.
Of course, this doesn't mean AI is replacing designers or video editors. Good branding, storytelling, and creative direction still matter.
For us at OSBR, it's about finding ways to combine human creativity with AI to make our workflows faster and more efficient.
And if we can already create promotional videos with just three to five prompts, we're pretty excited to see what's next.