Coupé

Inspiration

We looked at the wave of low-quality AI content flooding social media over the last months and saw a tool that could be used in a better context. Instead of contributing to the noise, we thought, what would it look like to take that same idea and redefine it in a way that's actually useful? That’s where Coupé comes in. It takes the concept of fast, automated content generation and reshapes it into a polished tool designed for websites and businesses.

What it does

Coupé takes in a website URL, studies the product, its features, and the surrounding context, then turns that understanding into a storyboard built to highlight the site in a way that feels deliberate and easy to follow. From there, it reads the user’s input to shape the video around their vision, whether that means a certain editing pace, a specific voiceover style, or a particular timeframe.

The platform includes more than 20 animation and editing styles, giving each demo more character and energy instead of making it feel like an overused template. The length is flexible too, with videos that can stretch up to seven minutes, so it works for both short introductions and more in-depth product walkthroughs. Coupé can also generate a voiceover that matches the tone and personality the user wants. A script can be provided manually, but if not, the platform can write one on its own using what it has learned from the website.

Beyond narration, Coupé adds sound effects throughout the edited sections to make the final video feel sharper, smoother, and more engaging for the viewer. Once everything is assembled, it delivers a demo video complete with editing and formatting that aligns with the user's prompt, making the process feel more streamlined without losing the sense of customization.

How we built it

We built Coupé as a pipeline, not a single model doing everything at once. The idea was to treat video creation more like a system of steps, where each part has a clear role and builds on the last. The program starts by pulling in the actual website using a headless browser. Instead of guessing what the site looks like, we work directly from its real structure, that way, everything that follows is based on what actually exists from the requested site. From there, an agent reads the site in pieces to map out design, the core idea, a voiceover script, sound direction, and a detailed plan for how the video should play out. Breaking it into parts kept the output consistent and easier to control on the user side. Remotion is then used where the system follows a set of patterns for timing, transitions, and layout so the videos feel intentional and not random like a basic template. Once the code is ready, it gets rendered into a final video, and after that, audio is added as its own step, with voiceover and sound effects aligned carefully so nothing feels off or cut short.

Challenges we ran into

We ran into a mix of challenges that required us to think beyond just writing code. Capturing the live structure of real websites was more difficult than expected, especially as layouts change dynamically and don’t always follow consistent patterns. On the rendering side, handling different viewports forced us to think carefully about how content adapts across different screen sizes, which added another level of complexity to our design process. At the same time, we had to actively control hallucinations in generated content as with any AI. Making sure our outputs stayed accurate rather than just sounding correct was more of a challenge than anticipated. Building longer-form content also introduced its own layer of difficulty, since maintaining coherence over time required more deliberate structure and validation. We also had to navigate practical limitations like request caps and rate limits, which meant optimizing how and when we made calls rather than relying on brute force. Designing the system to be scalable needed to think in terms of architecture rather than features, so that Coupé could handle growth with future updates without breaking down. Beyond the code itself, we faced challenges with hosting and deployment, where issues outside our control occasionally caused failures or downtime that stunted testing and performance.

Accomplishments that we're proud of

After working on Coupé we are proud of the quality of the videos generated and the level of customization the system provides. Our videos mimic the style of higher-end company demos, at a fraction of the cost and with a workflow that is accessible to virtually anyone. This allows individuals to produce more polished content without needing editing and marketing experience. In addition to that, the system adheres closely to user prompts, making sure that the final video reflects both the intended vision and the specifics of the product being showcased.

What we learned

What we learned from our work and research into this project is that automation becomes valuable when it is guided by people’s visions. People aren't looking for more content, they're looking for understanding, to see what something is, why it matters, and how it fits into their world without having to dig for it. Instead of generating videos quickly, the goal became to translate information into something meaningful. A good demo isn't visuals and voice, it's structure, pacing, and context all working together so the viewer can follow along. There's also a broader idea behind Coupé that we learned. Tools should not replace creativity, they should remove the obstacles around it. When the technical effort is reduced, people can focus more on what they actually want to say and how they want to present it. In that sense, the real benefit is not just saving time, but giving people a clearer way to communicate their ideas.

What's next for Coupé

Going forward, the focus will be on improving how accurately Coupé understands websites, making the outputs feel more natural, and giving users more control without complicating the generation experience in the process. The goal isn't to add more, it’s to improve what is already there so the tool feels more reliable and intentional every time it is used. We want to stay responsive to what people actually need to use Coupé as we intend. The experience at Hack Indy showed that ideas evolve quickly when they are tested with people and how much feedback there is to influence and improve a project. That means continuing to iterate, adjusting based on feedback, and rethinking parts of the system when something can be done better.