PanelRipper

PanelRipper: We watch the video so you don’t have to.

🎬 Inspiration

The internet churns out video memes faster than a hyper-caffeinated social network addict can swipe. Opening a dozen videos that demand minutes of my attention before I even know if they’re worth it - that’s torture. So I built a chainsaw for memes. This was imagined as an application that integrates everywhere, converting videos from every possible social platform or other source. But for starters, I slice Reddit meme videos that my friends (or are they really enemies?) constantly throw at me, into sheets that let me lol-and-forget, saving my attention for something useful.

Hero Comic Example

Example: 6-panel comic auto-generated from a video post.

🛠️ What it does

Feed PanelRipper a Reddit post link. About sixty seconds later you get:

A 6-panel comic that may substitute (well, some, at the moment) meme videos (and it could be even funnier than the original)
Your sanity saved, brainrot reduced.

Before and After Comparison

Before and after: Almost nailed it

🏗️ How we built it

Pipeline Diagram

A minimalist pipeline showing how video becomes a comic.

Frontend: Bolt.new to build from scratch around non-familiar React with Tailwind. User browser slices the video to save backend resources. In retrospect, this approach added unexpected complexity, see below.
Backend: Next.js running on Netlify, backed by Supabase. I slam the video frames and captions into LLM to get the comic script, then the image model produces the comic sheet in one shot.
Extras:
- Netlify handles static deployments.
- Supabase stores blobs, handles authentication, implements DB backend.
- Google Captcha, well, for captcha.

🧩 Challenges we ran into

Select video source for demo - nowadays most videos are hosted on services that do not give away anything.
Aimed at YouTube first - had to pivot to Reddit, as YouTube lacks APIs to officially get frames or transcripts.
Prevent platform resources from exhaustion by heavy video processing - off-loaded to the user browser, but smashed into the CORS issues (a day lost).
All platforms (Bolt, Netlify, Supabase) were unknown, my JS stack knowledge was obsolete.
Building with AI - re-starting from scratch 4 or 5 times, changing different approaches, learning how to work with Bolt in terms of git flow.
Building UI, actually - Bolt is tending to run ahead of the train, the project is growing insanely, Discuss and Approve is the way.
“Abusing” the infrastructure that was not meant for heavy video AI pipelines. There are ways, but I’ll move to proper infra ASAP.

🏅 Accomplishments that we're proud of

First of all, the project is live. The idea was growing in my head for years. Right place, right time.
Implemented an AI pipeline in 8 hours, as a balanced approach to meet the deadline. It is not even close to production grade, but produces funny and naive sheets.

📚 What we learned

Video processing on a serverless diet is not advised, but very possible.
“Vibe-coding” :) Non-vibe turned out to be a worse approach.
Know your technologies. For instance, Bolt tends to implement DB access open to authorized users (or everyone!), relying on RLS. While many fields in the table should be protected one way or another. Building without knowledge what is behind the scenes will lead to security, performance, or maybe even legal issues.
It takes more than expected, as always. With AI tools one can achieve fantastic results, but 80% of effort goes into actually delivering the thing.

🚀 What's next for PanelRipper

Enhance the image generation workflow. Current approach is held together with duct tape.
Community and feedback loop.
Localization.
Automated social publishing.
Give users more control - tone, interpretation, presentation, captioning, interactive editor, etc.
Cover more (much more) scenarios than meme videos - tutorials, presentations, podcasts, etc.
Add tools for creators to bake storyboards/previews.
Mobile app.
Integrate as a global video preview provider.