Don Quixote has a long record of defeating filmmakers. Cervantes' novel has been adapted many times, but it is also known for stalled and abandoned projects, with Terry Gilliam's repeated attempts being the best-known example. Guillermo Miranda and Javier de la Chica tried anyway, with AI as the main production tool. The result, La Triste Figura (The Sorrowful Figure), is a comedy-adventure built around the windmill episode.
La Triste Figura premiered in Madrid on 23 September. I was in the room at a small art cinema to gauge the audience reaction, something I've not seen before with an AI film, and it was interesting to see the comedy land and the dialogue resonate more than the confident action and style, which is something AI shorts have always handled well. In many ways, the film, which was co-produced by Magnific and made on the platform, feels warmer and more human than many 'AI-made' animations. So why?
First, story and script matter. De la Chica, a fan of the novel, says he's spent years thinking about what an adaptation could look like, and these AI tools finally let them try. Miranda says their earlier films drew on local Spanish culture, and that choosing a story everyone knows lets them add their own tone without explaining the context.
AI can feel like 'multiplayer' filmmaking
What interested me most wasn't the imagery but how the work is organised and how they ensured a 'human' input underpinned the use of AI tools. Both come from animation, where production is split into departments, from storyboards and design through animation, lighting, rendering and editing, and artists often stay within one of them. On this film, a small team worked across those stages more loosely: ten people according to the official film notes, though Miranda says it's around a dozen and up to 30 at the peak of production. The directors worked from different cities in a shared project, watching each other's generations in real time.
"We've been working with AI for a long time," de la Chica says. He began with early tools around 2019 and now describes AI as the pair's medium. Their aim is to make AI production "a multiplayer game", he says, pointing to how everyone in the team can have creative control, because the tools tend to feel built for "one-man armies". For them, cinema has always been collective creation, and Miranda says they are trying to find the balance.
This is the first of their films to include a storyboard artist and a composer. Víctor Ponce drew the storyboards by hand and is also credited as assistant director. The screenwriter Álex Rey helped with the jokes, and Arturo Liberos and Juan Rabadán voiced Quixote and Sancho. "We're trying to understand what the role of these artists is in AI production," Miranda says. One answer is for the storyboard artist to act more like an assistant director, shaping ideas rather than only drafting thumbnails.
Their process goes beyond typing prompts because, before generating anything, they built a production design with turnarounds, expression sheets, costume tests, and combat poses for the giants, plus a group image of Quixote, Sancho, Rocinante, and the donkey to lock in proportions. They then generated shots as video from those references and prompts, mostly written in Spanish, mainly with Seedance 2.5, with characters and assets built in Seedream 5 Pro. Miranda calls the process a hybrid, though still AI-built. De la Chica says the model follows instructions closely, but there is still an element of chaos, which they sometimes use in the edit to find new beats. According to the press notes, the work took around 1,200 generations for about 120 shots in the final cut. Because Quixote's design is asymmetrical, they couldn't flip a good shot, and a duplicated sword or a cape that switched sides meant starting again.
Human craft is still visible, with early AI voices scrapped because they stumbled on archaisms and failed to deliver the weight the characters needed. For the film's central shot, in which Quixote pushes himself up on his sword, de la Chica acted out the movement on the floor himself, as animators do for reference. "In AI, everything is about intuition," he says, and animators have spent years building that instinct. Miranda says audiences still like to see and hear humans, which is partly why they use voice actors.
For animators used to everything being calculated to the millimetre, this took time to get used to. Miranda says they had to "unlearn" some instincts, because comparing the results with their animation work was frustrating at first. He teaches animation and tells students that limits push creativity. With AI, he says, "the limits are moving to a different place." He describes a motto of trying to do things "almost good": finding the edge of what the tools can do and pushing there, because anything now looks possible; he says the question becomes, "Does the story really need it?"
They don't claim that using AI for animation matches a traditional studio workflow. "This is not a technical showcase of animation," Miranda says, and he calls the film a prototype, meant to make you laugh or feel something. He describes the results as far from animation and "not the same thing," and says industry standards still have a way to go, though for entertainment it is "getting quite ready." It isn't right for every field: Miranda, who also works in medical animation, says it makes no sense there today because accuracy matters.
Nor do they see it as cheap or open to everyone, as many AI supporters have argued. De la Chica says it is cheaper than traditional animation, but you still need funding and a team, and it is "not yet a full democracy" of the medium. What it offers, he says, is a way to take risks in an industry where high costs mean the projects that get made tend to be established properties. Miranda compares it to the home studio in music, but says he isn't sure whether that is a good or bad thing.
Both are open about the unease that can come with AI being mentioned and used, as Miranda says some friends in the industry are unhappy with what they're doing, while others are interested but afraid to say so because of the polarisation. He says there are "many shadows" around the technology that they are not hiding, and that if a new industry is being built, they want a say in how it works and what it produces. "We are trying to find the grey area, open the conversation," he says, explaining that they also don't know whether audiences will accept AI-made films. Right now, it all feels like an experiment.
On jobs, de la Chica expects studios with hundreds of artists to employ fewer people. He hopes the volume of content and the number of smaller studios will grow, but he doesn't claim to know how things will pan out. "We're in the fog a little bit," Miranda says.
De la Chica's favourite part of the film is a roughly five-minute comic dialogue sequence with windmills turning in the background, which takes up nearly half the running time. Some of the film's most visually complex shots were easy by comparison, he says, but those dialogue-heavy, comedic scenes were "a big challenge […], and we're proud of that", because AI is trained to do effects shots, big stylistic action sequences, "but when you try to put a conversation, a five-minute conversation, four characters in the same place with windmills working in the background, it's extremely complex".
He admits he might choose more wisely if he started again. Keeping characters consistent and making a scene work as storytelling was the hard part. The windmills were a problem of their own: they resisted turning at a constant speed without warping, so the directors gave up on physical realism in favour of comic effect.
Fixing problems is odd too when using AI, and it's where the gaps begin to show. De la Chica says small changes are hard, but a whole new sequence is possible, and the film's hardest problem ended up adding about a minute and a half to the run time in order to solve it. Comparing the cost and work involved in AI fixes to traditional GC, Miranda says a subtle tweak costs about the same as a shot full of hair, liquids and fire, or "one guy in a room with nothing," so why not change everything rather than one thing?
For the pair, working across every stage is a different kind of enjoyment than a traditional CG animation workflow. De la Chica says they miss parts of the traditional process, but that artists in a 400-person studio can spend a production focused on their own shots and only learn what the film is about at the premiere.
Asked whether they want to make a "pure AI" film, one with no human actors, traditional storyboards, script or editing, Miranda says they don't focus on AI for its own sake and want to bring in people from the industry where they can; the aim is simply to tell the story. It's revealing because they both view AI as a way to tell stories that enables more direct authorship, not an end in and of itself.
Watch La Triste Figura (The Sorrowful Figure), available on Magnific.
