There is a moment in Obsession where Nikki stands in the dark corner of Bear’s bedroom and something about her is wrong. Reviewers kept reporting the same experience, which is the useful part: you can see that the face is off, and you cannot name which bit of it moved. The Post Athens called the film a masterclass in subtlety and put that shot down to special-effects makeup. Most of the write-ups that noticed it at all said some version of the same thing — practical work, not the computer.
Half right, as it turns out.
Curry Barker’s debut premiered at TIFF’s Midnight Madness in 2025, got picked up by Focus Features with Blumhouse’s Jason Blum on as an executive producer, and opened in US theatres on 15 May 2026 in a little over 2,600 screens. Michael Johnston plays Bear, the romantic who breaks a novelty charm called the One Wish Willow and wishes his childhood friend would love him more than anyone alive. Inde Navarrette plays Nikki, who gets the wish. Six months after release, a befores & afters interview published today with production visual effects supervisor Peter Timberlake sets out what his studio, Queen, actually did to her face — a mix of practical make-up and 2D compositing, built around an in-house toolset they call Plastik, aimed at what he describes as a deliberate uncanny valley. The enhancements are, in his words, of the invisible variety: subtle shifts to the actor’s features and nothing more.
The design came off a phone
The makeup it was built on has a stranger lineage. Bloody Disgusting reports that Nikki’s look was drawn from an uncanny-valley trick popularised on TikTok by a beauty creator, Emilia Blarth, around 2023, with key makeup artist Kate Oja sharing the design images. So the scariest face in a studio horror release traces back to amateurs working out, on a phone camera and with a shelf of drugstore products, which small deviations in a human face make a viewer uneasy.
The makeup department earns more credit for that, not less. Figuring out which deviations is the entire problem, and it is a perceptual question before it is a technical one. A beauty creator chasing views on a short vertical video is running the same experiment a facial rig gets run through in a review session, with a faster feedback loop and a larger sample.
Why the top-up stayed in 2D
Here is where it gets interesting for anyone who builds faces for a living.
The obvious route for “make this actress look subtly inhuman” is a 3D one: scan the performer, build the head, solve the performance onto it, render, comp. It is the route a lot of films take, and it buys you total control. Every millimetre of the face becomes a parameter you can dial.
It also throws the take away. Once the head is rebuilt, every flicker the actor did on the day has to be recovered through a solve, and solves lose exactly the things that make a face read as alive — the asymmetric micro-expressions, the small timing offsets between one side of the mouth and the other, the eyes arriving a frame before the brow. On a film where the lead performance is the engine, and most reviews of Obsession agree that Navarrette’s is, spending money to rebuild the thing you hired her for is a poor trade.
A 2D warp keeps her on screen. The cost lands elsewhere, and it is real work:
- Tracking skin that deforms. A face is not a rigid object, so there is no single transform to lock to. You are tracking regions that stretch and occlude each other.
- Warping without tearing, frame over frame, while the head moves and the light on it changes.
- Matching grain, lens softness and the falloff at the edge of the warp, which is where invisible work usually stops being invisible.
Do all that well and nobody can point at it. Which is the trap.
The credit problem
Work this good gets attributed to the makeup chair by default, because makeup is the explanation a viewer already has available. The comp pass has no visible signature, so it collects no credit, and for six months the public record of that shot said “practical effect” with nothing after it.
The blog has watched the same accounting run the other way. When Clayface landed in October, critics agreed about the prosthetics and split on everything the computer had to finish — praise to the physical department, blame to the digital one, from the same screening. Both halves of that habit come from the same place. Makeup is a thing in the room; compositing is an absence of a seam, and absences do not get reviewed.
Set it beside the other way this problem gets solved. Danny Boyle’s Ink put an AI Thatcher over a real performer for about thirty seconds — an impersonator played the scene and the face went on top of her. That approach announces itself, gets argued about, and generates a week of coverage about what it means. Queen’s approach generated none, and did more for the film.
The rest of the sheet
Two other things in Timberlake’s account are worth a note. The interview describes a previs approach he calls unusual, which is itself a flag — previs on a first-time director’s horror feature at this budget is not a given, and it tends to show up where a filmmaker has a specific effect he cannot afford to discover on the day. And there is a final-scene shot he calls completely undetectable, which he has now volunteered in public, presumably because nobody would ever have found it.
I have not seen the breakdown reel, only the interview and the coverage. Judge the rest when the frames turn up. But the pattern is already clear enough, and it is the one worth stealing: let the department that can solve the perceptual problem solve it in front of the lens, then spend the digital budget on four pixels of nudge rather than a whole new head.