About
How Cinemoodly works
Most film search starts with a title. This one starts with a description: say what you feel like watching, in your own words, and the search returns films with the same tone — even when none of your words appear in them.
«a dark seaside mystery, rain and an old secret»
How this differs from ordinary search
Keyword search checks for overlap: does the text about the film contain the same words as your query. That works beautifully when you already know what you are after — a title, a director, a year.
But most of the time you do not know the title; you know a state of mind: “something warm and quiet”, “beautiful but unsettling”, “a film to cry to”. There is no word overlap there, and there cannot be — the film you want is described in completely different words. Meaning-based search closes that gap: it compares the sense of the whole phrase rather than its letters.
A side effect: synonyms, paraphrases and typos stop being a problem, and the language of your query does not have to match the language of the description.
What actually happens
There are no keyword lists inside and no hand-written rules of the “if someone asks X, show Y” kind. There are four steps, the same for every film and every query.
- One text is assembled per filmTitle and year, genres, themes and tags, franchise, director and cast, the plot summary — plus a separate description of the film’s atmosphere written by a language model from that same data. English and Russian texts are built separately and never mixed.
- The text becomes a list of numbersA neural model turns the text into a vector — 3072 numbers that act as coordinates of meaning. The key property: texts that mean similar things get similar vectors, even when written in different words or different languages.
- Your query takes the same pathThe phrase you type is turned into a vector by the same model. From that moment the query and the films live in one space and can be compared directly.
- The nearest ones winThe database picks the films whose vectors sit closest to the query — that is what “similar in meaning” means here. You can then reorder the results by rating, popularity or year, and narrow them by genre, theme, actor, years or IMDb score.
Three search modes
Meaning-based search is strongest on descriptions and moods, but it is not the right tool for everything. Hence three modes, switchable right at the search box.
- By meaning
- Everything described above. Use it when you are describing a film rather than naming it.
- Hybrid
- Meaning plus exact words — useful when the query contains a name, a franchise or a rare term that must not be lost.
- By words
- Plain text search. The shortest path when you know the exact title.
What it cannot do
This is a way to find something similar, not an answer to a question. No query has one correct result list: two phrasings of the same idea will return different films, and that is the method working as intended rather than failing.
The atmosphere description attached to each film is written by a language model from metadata. It can be wrong in the details, which is why those texts are labelled on the film page itself.
The search can only reach what is in the catalogue. If a film is not there, no phrasing will surface it. The catalogue keeps growing, but it will never hold every film ever made.
Where the data comes from
Descriptions, posters, genres, themes and credits come from TMDB. Ratings and vote counts come from the IMDb non-commercial datasets. We do not rewrite anyone: film texts are used as they are, and missing translations are produced by machine translation — those are labelled too.