Can AI Generate a Professional Book Trailer?
Not a day goes by that I don't noodle with artificial intelligence in some way. For example, last June, I used NotebookLM to generate two podcasts based on a chapter of Low-Code/No-Code. It also churned out a respectable book trailer based on its synopsis. The results impressed me, but AI is rapidly advancing, I wondered:
How much have AI multi-modal capabilities improved the last 16 months?
I decided to answer this question by giving a few new open-source video-creation tools a spin in Claude Code. Could they generate compelling book trailer for my latest effort, The Nine?
Let's find out.
Brag
I began with Brag. As expected, installing it was easy:
/plugin marketplace add latent-spaces/brag
/plugin install brag@bragBy way of background, I'm on Anthropic's Pro plan. To minimize the tokens required to create the trailer, I ultimately went with Brag's slim version. I then entered the following prompt in Claude Code:
/brag-slim on www.philsimon.com/book-the-nine.
Maybe a minute later, the following Brag/Claude offspring appeared on my computer:
Brag First Attempt
Impressively, Brag independently provided the assets it used to create the video:
Brag also provided an editable markdown file of its video timeline with timestamps:
Technical Details
๐ฒ๐๐๐๐๐๐๐ข ๐๐ ๐ฒ๐๐๐๐๐ ๐ฒ๐๐๐
Built complete motion graphics pipeline using page scraping, PIL font rendering, scene-based animation with easing functions, and procedural audio synthesis (44.1kHz stereo, harmonic oscillators, feedback comb-filter reverb). Created 45-second H.264/AAC video (1920ร1080, 30fps) with 8-scene timeline (hook, companies, forces, book, Kirkus quote, investor praise, outro) using only math and wave modules for audio. Produced video assets: brag.mp4 (21.5s original, extended to 45s), poster thumbnail, narrative breakdown, and distribution copy sourced entirely from published page text.
Initial Results
Was the Brag trailer perfect? Not even close. The first few seconds of the video don't make a hell of a lot of sense. Brag truncated a sentence from the book's description:
Executives and managers at these organizations are trying to turn back the clock to 2019.
Minor gaffes aside, Brag in 2026 produced a better video than NotebookLM did in June 2025.
A Second Attempt
How would Brag do when I prompted it to expand its original trailer?
See for yourself:
Brag Second Attempt
My efforts ran me a modest number of tokens:
Brag is not the only GitHub repos out there. Far from it. I knew that other ones were worth exploring.
HyperFrames
Ever curious, I next decided to kick the tires on HyperFrames. It walked me through some prompts. Soon a freakin' storyboard appeared. I could edit it before rendering the video.
I knew other skills were worth exploring.
In a word, wow.
HyperFrames even displayed the rendered video a local, browser-based editor with different tracks:
Here's its first version:
Mind blown. I still prefer the original, human-generated 2023 video. Here's the rub, though: the 2023 trailer cost $1,000.
Final 2022 Trailer
I suspect that a proper videographer could take both my Brag and HyperFrames videos and add a little spice. Maybe I'd drop $200 to get something comparable to the 2023 trailer.
Phil Simon
Simon Says
I began this post by posing simple question. Answering it leaves me confident of things. First, the ability to quickly and easily create cheap content means far more AI slop. Second, a year from now, AI-generated videos will put my efforts today with Brag and HyperFrames to shame. Even when the AI bubble crashes, plenty of open-source tools will remain and flourish.
Feedback
What say you?
Member discussion