Unlike a Google search, Flim,ai’s engine can source film stills by theme, colour scheme, the number of people in the frame, and yes, an art-related scene like Mexican artist Frida Kahlo’s ‘cameo’ in the 2017 film Coco. (Image courtesy Pixar) Exclusive
Unlike a Google search, Flim,ai’s engine can source film stills by theme, colour scheme, the number of people in the frame, and yes, an art-related scene like Mexican artist Frida Kahlo’s ‘cameo’ in the 2017 film Coco. (Image courtesy Pixar)

And now, a bot that knows the exact movie you’re thinking about

Flim, a new AI-driven search engine, can “read” film stills, picking up on emotion, colour, mood and style. Could it help with your vague movie memories or should you be afraid, very afraid?
UPDATED ON JUN 04, 2021 09:29 PM IST

For movie buffs, life just got easier (and a whole lot more complicated). It used to be that if you were trying to track down a move by title, release year or a member of the cast and crew, you’d head to the Internet Movie Database (IMDb). To measure Cara Delevingne’s privilege, you’d look up her illustrious family tree on Wikipedia. To track a film poster, you’d look through the million uploads on CineMaterial. Rotten Tomatoes collates reviews. The Complete Index to World Film lives up to its title. There’s even a dedicated Internet Movie Firearms Database.

But what if all you had was a hazy memory of a scene, a half-remembered shot? You know it had a puppeteer. Perhaps putting on an inappropriate show on the street. A nun may have been involved. With keywords this vague, Google is of little use.

Enter Flim.ai. The algorithm-driven searchable archive is just under a year old, and is already the world’s largest database of film-related images. Flim’s algorithm “reads” through more than 3 lakh high-definition images from movies, documentaries, anime, advertising and music videos. The bot can detect colour palette, genre (from the film’s metadata) and aspect ratio. It can also precisely identify dog-walkers, ham sandwiches, red-lipped Asian women, animated metropolises and other visual cues. It’s pattern identification, facial recognition, data mining and pop-culture cross-analysis all rolled into one.

For those vague memories, it’s perfect. Flim.ai immediately connected our vague keywords to a minor scene from the 1999 indie film Being John Malkovich (it did indeed feature a puppet of a nun).

Wes Anderson’s 2014 film The Grand Budapest Hotel follows a tight visual palette. You can look for film stills by colour on Flim.ai to find other movies that match this hue. (Image courtesy Fox Searchlight Pictures)
Wes Anderson’s 2014 film The Grand Budapest Hotel follows a tight visual palette. You can look for film stills by colour on Flim.ai to find other movies that match this hue. (Image courtesy Fox Searchlight Pictures)

The site is the work of Dan Perez, 35, and Victor de Casteja, 33, who met while they were studying video photography in Paris in 2009. They’ve since produced music videos and commercials, and worked in fashion, advertising and art. “I created Flim because I am mad about movies,” says Perez. “As a student I watched a lot of films and spent a lot of time screenshotting them for inspiration for my own videos.”

Some 40,000 of those shots formed the basis of an early experimental site – Perez put them online, allowing visitors to share their screenshots too. It made him realise that there was a market for what he calls “iconographic searching”.

Still in its beta phase — it’s supported by French incubators Paris&Co and Belle de Mai — the site is already a hit with creative folks.

“It’s a handy tool for when you’re looking for something specific,” says Sara da Costa, a Mumbai-based ad-film director. Clients rarely share the same pop-culture references as creative people. They struggle to visualise, say, the slow-motion shower of rose petals from American Beauty or the marigold fantasy from Monsoon Wedding. “So being able to have a handy visual reference is useful,” she says.

Where Flim.ai flounders is where AI everywhere is currently floundering. Machines lack the ability to make connections that mimic the human experience. When da Costa used the site to look for green rooms, the AI showed her green-walled rooms rather than, say, a backstage scene from Birdman or Black Swan. “Ultimately, my memory and that of my assistant director are more reliable,” she says.

Pride and Prejudice and Zombies (2016). Looking for a specific period drama, battle montage or even a movie with heavy drapery? Flim.ai can match those keywords with reasonably precise results. (Image courtesy Lionsgate)
Pride and Prejudice and Zombies (2016). Looking for a specific period drama, battle montage or even a movie with heavy drapery? Flim.ai can match those keywords with reasonably precise results. (Image courtesy Lionsgate)

For now, there also aren’t enough stills from Indian films. The catalogue doesn’t extend beyond a handful of contemporary hits such as Devdas and Kuch Kuch Hota Hai. “It’s a drop in the ocean of Indian movies,” Perez admits. But there are plans to collaborate with distributors and add world cinema to the site. “The aesthetic in Indian movies is very strong, a visual treasure that we lack on Flim.”

Meanwhile, prepare to wade through a glut of anime shots when you search with certain keywords (“noodles”, “hope”, “alone”). And while you can search within a specific film (just the rain scenes in Parasite, for instance), you’re unlikely discover a gem through keywords alone.

For anyone in the visual arts who’s struggled to build a moodboard, the idea of a bot being able to put one together sounds both liberating and limiting. Interior design assistant Prakrit Kumar tried using the site for ideas for a commission he’s working on — the bedroom of a teenage girl. “I saw all these American stereotypes – muted pink, ivory, a photo wall, fairy lights – the very things I’d avoid as a designer,” he says. “In the creative field, a machine can only go so far, the human must ultimately take the standard idea a step forward.”

Hit rewind: Online archives for film buffs

BFI National Archive: The British Film Institute is a bonanza for anyone looking for old reels and films. It looks after one of the world’s largest most important film and television repositories. On YouTube, their playlist titled India on Film: 1899-1947 contains more than 100 clips of an India you may not even recognise. It’s time-travel without the nosebleeds.

FSUE Mosfilm Cinema Concern: Films from the largest and oldest film company in the Russian Federation. Switch to auto-translate on the YouTube channel and settle in with Yuri Ozerov’s sweeping two-part Battle of Moscow (1985), or Yevgeny Karelov’s 1976 mini-series Two Captains, based on the popular children’s novel about a young man’s search for a lost Arctic expedition. There’s plenty of Eisenstein and Tarkovsky too.

Center for Home Movies: What happened to all those amateur films Americans shot with their beloved (and clunky) Handycams? This organisation works to collect, catalogue and preserve them as cultural heritage. The archive contains scenes from everyday life, and is a riveting record of what was deemed worthy of filming by regular folks empowered by a camera. A selection of the films is available for viewing on centerforhomemovies.org.

British Pathé: Quite simply the finest newsreel archive in the world. The YouTube channel features a fraction of their 85,000 films of historical and cultural significance — clips of British royal weddings, war preparation, life in the colonies (including India) and new industry. Also part of the archive is the Reuters historical collection, 1.3 lakh clips from news agencies, some more than a century old.

Korean Classic Film: Put your subtitles on to watch more than 125 films uploaded in HD on YouTube. Black-and-white dramas from the 1960s, some very melodramatic family sagas, and grown-up films that require you to sign in and confirm your age. And, of course, love stories that have much in common with today’s K-soaps.

Please sign in to continue reading

  • Get access to exclusive articles, newsletters, alerts and recommendations
  • Read, share and save articles of enduring value
SHARE THIS ARTICLE ON
Close
SHARE
Story Saved
OPEN APP