SceneThat — AI Movie Discovery Platform

SceneThat is a concept web platform that helps people identify movies and TV series from scenes instead of titles. Paste a clip, upload a screenshot, share a social link, or just describe a scene in plain text — and it returns the most likely match.

The idea came from a real, specific frustration: you scroll past a great movie clip on TikTok or X with no title, no context, no way to find the source. A moment of curiosity hits a dead end. SceneThat was designed to close that gap — and to make AI-driven discovery feel conversational and human rather than technical.

People discover movies through fragments — a short clip, a screenshot, a scene shared in a group chat, a vague memory of a moment. But traditional search expects you to already know the title, the actors, or the right keywords. When that information is missing, discovery breaks down.

The question users are actually left with is simple: "What movie is this from?" Everything in SceneThat is built to answer it.

Design a scene-first experience that:

• Works from incomplete or vague input
• Supports images, video, links, and text
• Feels fast, confident, and conversational
• Guides users from discovery through to where to watch

Instrument Serif for headers — it gives the product a cinematic, editorial feel that fits a movie tool — paired with Overused Grotesk for body and UI. A purple-led palette (#7C35E7 primary) against near-black section backgrounds keeps it feeling modern and AI-native without leaning on the usual chatbot blue.

Instead of a traditional search bar, SceneThat opens with a chat-style input familiar to anyone who's used a modern AI tool. It accepts four kinds of input — text description, image upload, video clip, and social links (TikTok, X, and the rest).

By default the input carries a colourful gradient border to draw attention as you land. Once you start typing, it reverts to a calm grey field for readability. Add a description and the plus icon reveals options to attach an image or video — so the interface stays simple until you need more from it.

Hit search and the input becomes the first message in a chat between you and SceneThat's AI. The AI responds confidently but concisely — the movie name, the official poster for instant recognition, and a short description.

The goal at this moment is trust. The user should feel: "Yes — this thing understood what I meant." To keep that momentum, the interface deliberately doesn't stop to ask "is this correct?" Instead it offers lightweight actions — continue, refine, or reject — so the flow never turns into an interrogation.

When the user disagrees. If someone says the match is wrong, the AI adapts immediately. No defending the suggestion, no restating confidence. It acknowledges the miss and shifts into refinement mode, prompting the user to narrow things down in plain, non-technical ways — faster or slower, indoors or outdoors, action or conversation. It's positioned as a helper, not an authority, which is what builds trust and defuses frustration.

Not every scene description is clear, and the system is honest about that. Instead of guessing, it surfaces a small set of ranked possibilities and invites the user to guide the next step — pick the one that feels right, or add another detail. It reframes uncertainty as collaboration rather than failure, which respects how memory actually works: fragmented, approximate, and often emotional rather than precise.

Where to watch lives right on the movie card. Rather than opening a full-screen modal or navigating away, the card expands inline to reveal streaming options — keeping the user anchored in the conversation they started. High intent, no broken flow.

The detail view. An "i" icon on the card opens a modal inside the chat, not a new page — reinforcing that discovery is all one flow. It gives depth without overload: title, aggregate ratings, a short spoiler-free description that sets tone, similar movies, and trailers. It answers the immediate question — "Is this worth my time?"

Beyond the core loop, SceneThat is a full product:

Navigation. A persistent profile entry at the bottom (avatar, name, plan) doubles as a global nav trigger. Tapping it opens a lightweight modal to Watchlist, Communities, Settings, and more — branching out without a heavy sidebar or top-nav pulling focus.

Watchlist. A memory layer. Saved movies organise into custom folders by mood or intent (Favourites, Mind-Bending, Seen on TikTok), each showing recency and count, doubling as a soft discovery surface.

Communities. Framed as trusted extensions of discovery (Reddit, Telegram, Discord, X), on the honest premise that movie discovery is social — people love talking about what they've watched as much as watching it.

Explore. For the different mindset: "I don't have a scene in mind, I just want something good to watch." Categories and scrollable rows (Top Picks, Recently Added, Trending) for low-pressure browsing.

Kept intentionally lightweight — Google, Apple, or email, with social sign-in prioritised visually. No mandatory preferences or onboarding forms; personalisation happens gradually through use. Reduces friction and encourages exploration before commitment.

This project grew out of a familiar frustration: stumbling on a great movie clip online and having absolutely no idea what it's from. Existing platforms attempt to solve this, but many fall short when the input is vague, fragmented, or purely descriptive.

SceneThat explores how AI can step in as a more flexible collaborator — making sense of scenes, screenshots, links, and loose descriptions the way a human would, not the way a rigid search box expects. Instead of demanding perfect keywords, it's designed to listen, adapt, and respond with confidence when it can, or explore alongside the user when it can't — turning "Where is this movie from?" moments into clear, satisfying answers.