Stitch

Inspiration

Stitch started from a simple frustration: video editing is powerful, but it’s still unnecessarily hard to use. Even modern editors expect users to think in timelines, tracks, shortcuts, and effects panels. For beginners, that learning curve is intimidating, and for experienced creators, it often interrupts creative flow. We wanted to explore what video editing would look like if creators could work the same way they communicate with another person—by simply describing what they want to happen.

What it does

Stitch is a web-based AI-powered video editing platform that combines a traditional timeline editor with a natural language assistant called Lilo Agent. Users can edit videos manually using a familiar timeline interface, or they can describe edits in plain language, such as trimming clips, rearranging scenes, adding audio, or generating transitions. Instead of just suggesting changes, the AI directly edits the timeline while fully respecting its structure and constraints. This allows creators to move fluidly between hands-on editing and AI-assisted workflows without losing control.

How we built it

The platform is built as a modern full-stack web application using Next.js with the App Router, React 19, TypeScript, and Tailwind CSS. The timeline editor relies on Zustand for state management, with dedicated stores for video clips, audio layers, selections, clipboard actions, and undo/redo history. On the backend, we use Next.js API routes with Prisma and PostgreSQL, while authentication and file storage are handled through Supabase.

AI features are deeply integrated throughout the system. Uploaded videos are automatically indexed using Twelve Labs, enabling semantic search across visual content, audio, and transcripts. Lilo Agent is powered by large language models that reason over timeline state and invoke structured editing tools. Text-to-speech audio is generated through ElevenLabs, and Stitch already supports AI-generated video transitions using VEO. These transitions are created by extracting the last frame of a clip and the first frame of the next clip, then generating a smooth, context-aware transition video that is automatically inserted into the timeline.

Challenges we ran into

One of the biggest challenges was translating vague human intent into precise, deterministic edits. Video timelines have strict rules; clips cannot overlap incorrectly, trims must be exact, and audio layers require careful depth management. Another challenge was building trust in AI-driven edits. To address this, every timeline modification, whether initiated by the user or by the AI, flows through the same command system with full undo and redo support. This makes AI actions feel transparent, safe, and easy to experiment with.

Accomplishments that we're proud of

We’re especially proud of building a fully functional timeline editor in the browser and integrating an AI agent that performs real edits instead of offering passive suggestions. The combination of semantic video search, structured AI tooling, and reversible timeline commands resulted in an editing experience that feels both powerful and approachable. Seeing AI-generated transitions blend seamlessly into the timeline alongside manual edits was a particularly rewarding milestone.

What we learned

Through this project, we learned that AI is most effective when it operates within well-defined constraints. Giving the model explicit tools and access to structured state produced more reliable and controllable behavior than free-form generation. We also gained valuable experience designing complex interactive systems, coordinating frontend and backend timelines, and orchestrating multiple AI services into a cohesive creative workflow.

What's next for stitch

Looking ahead, the next steps for Stitch focus on refinement and scale rather than core functionality. We plan to improve preference learning, so Lilo Agent adapts to individual editing styles over time, making its edits feel more personal and consistent. We also want to expand AI-assisted workflows beyond single commands into multi-step edits, where users can describe higher-level creative goals and review staged changes before applying them. Longer term, we see Stitch evolving into a full AI-assisted creative workspace, one that helps creators move from raw footage to polished output faster, without sacrificing precision or creative control.