Hell Grind: The 95-Minute AI Feature Cannes 2026 Says It Never Screened

AI

by Nino LeitnerMay 28^th, 2026

Subscribe to News - Industry

Higgsfield AI used the gravity of the 79th Festival de Cannes to unveil Hell Grind, a 95-minute generative sci-fi feature its team of 15 built in 14 days for under $500,000. The film never screened in the official program, something the festival itself went out of its way to state publicly, yet the pipeline behind it is the part working filmmakers should actually be looking at.

Let us start with the detail that has been widely garbled. Higgsfield held a private industry preview of Hell Grind on the Vieux Port on May 16, followed by a fuller screening at the Cinéma Olympia on May 21. The Olympia is a working commercial cinema in the town of Cannes, not a Festival de Cannes venue, and both events sat within the broader Marché du Film orbit rather than the Official Selection. When the Wall Street Journal reported the film as debuting at the festival, Cannes pushed back directly, confirming to Futurism that Hell Grind was not part of the official program and had instead been presented during an industry event organized by third parties. CEO Alex Mashrabov has happily blurred that line since, describing the work on LinkedIn as having “premiered in Cannes” and casting the city as the room where new cinema gets legitimized.

What Hell Grind actually is

Hell Grind is a sci-fi action-fantasy heist directed by Kazakh filmmaker Aitore Zholdaskali, whose 2025 feature Sicko played Rotterdam and SXSW London, and co-written with Adilkhan Yerzhanov, a filmmaker who has twice appeared in Cannes official programs. The logline follows four street thieves, Roco, Lulu, Jax and Rein, who open a portal to the underworld after a botched heist and chase it through a Tibetan temple and feudal Japan. Higgsfield pitches the tone as fantasy as tragedy, action as grief. UK finance and production outfit Goldfinch is attached as a studio partner.

One wrinkle is worth flagging early. While the press materials consistently describe a 95-minute feature, Higgsfield is also releasing the work on its own platform as an episodic series of roughly 22 to 25 minute installments, so the feature-film framing and the original-series framing coexist a little uneasily.

How 15 people generated a feature in two weeks

This is the part that matters for our readership. Production ran 14 days with a 15-person team of directors, DPs and editors working mostly in person from Almaty, Kazakhstan. The stack was Higgsfield’s own: Soul Cinema for cinematic image generation, Soul Cast for character and face consistency, Cinema Studio 3.5 as the multi-shot workspace with its 1,296 virtual lenses, and the DoP image-to-video model, all orchestrated on top of ByteDance’s Dreamina-Seedance 2.0 as the primary video backbone with Google Veo 3 also in the mix.

The numbers behind the curation are the real headline. The first 25-minute segment alone required 16,181 generations to land 253 final shots, a ratio of roughly 64 to 1. Prompts ran around 3,000 words each and produced about 15 seconds of footage per attempt. To fight the floaty, overlit look that betrays most generative video, the team wrote explicit physics into the prompts, instructing the models to respect gravity and inertia and adding phrases such as “no floating props,” along with natural-light prefixes intended to kill the telltale AI sheen. Compute ran on Nebius and CoreWeave rather than the major hyperscalers, a choice Mashrabov says kept costs down. Of the sub-$500,000 budget, roughly $400,000 went to compute.

Higgsfield’s claim that a comparable traditional action-fantasy would run about $50 million is its own figure, repeated across outlets but not independently verified, and the comparison is strained given that the AI version carries no actors, locations, insurance, transport or crew in its cost base. For a grounded, practitioner’s view of this kind of pipeline rather than a marketing number, cinematographer Drew Geraci’s continually updated MZed course Directing the Future: Ethical AI Video for Filmmakers works through real hybrid workflows that blend generated elements with live footage. It has also added dedicated modules on Higgsfield itself, which we covered when Drew first explored the platform and again when he built pre-production mood boards and wardrobe looks with it for a real shoot.

A $1.3 billion company hunting its next market

To read Hell Grind correctly you have to see it as a sales reel. Higgsfield was founded in 2023 by Mashrabov, previously head of generative AI at Snap after Snap acquired his AI Factory startup in 2020, and Kazakh AI researcher Yerzat Dulat. The company is headquartered in San Francisco with a large engineering presence in Kazakhstan, and it has raised an $8 million seed plus a Series A that now totals $130 million at a $1.3 billion valuation, making it Kazakhstan’s first unicorn. Revenue intelligence firm Sacra estimated the company hit a $300 million annualized run rate in February 2026, and Higgsfield’s own materials claim 25 million users.

Most of that revenue still comes from social-media marketers buying credits to make short-form ad content, which is exactly why a 95-minute feature matters to the company. Hell Grind is the most visible attempt yet to push the same stack into long-form cinema and, in Mashrabov’s framing, to show studios that the infrastructure now exists to execute complex visions at a fraction of traditional cost. The film is the pitch; Hollywood is the customer.

Cannes drew its own line

The festival has not been shy about where it stands. At the April 9 press conference unveiling the Official Selection, president Iris Knobloch and general delegate Thierry Frémaux set out a curatorial exclusion of films primarily driven by generative AI from the Competition, a stance applied in the selection of the 21-film lineup even though it does not yet appear as a codified rule. Knobloch put it plainly, saying a movie is not just a patchwork of data but a personal vision, and that AI can imitate but will never feel. On opening night, during the 20th-anniversary screening of Pan’s Labyrinth, Guillermo del Toro drew the loudest line of all with a blunt “Fuck AI!” aimed at anyone arguing that art can be made with an app; Frémaux, on stage moments later, called it the first political statement of the festival.

The position is not without its contradictions, and they are worth naming. Meta now holds the festival’s headline tech-sponsor slot, and Steven Soderbergh’s Lennon documentary, built partly with Meta’s AI tools, screened as a Special Screening, a tension we covered in our piece on how that film divided Cannes 2026. The festival’s discomfort, in other words, is real but selective.

The market it is really aimed at

Hell Grind arrives in a generative-video field that has reshuffled fast. OpenAI discontinued Sora in April 2026, leaving Google’s Veo, Kuaishou’s Kling, ByteDance’s Seedance and Runway’s Gen-4.5 as the surviving frontier, a landscape we mapped when Kling 3.0 launched. Higgsfield’s pitch is to sit on top of all of them as an orchestration layer that adds character consistency, camera-language presets and those 1,296 virtual lenses. At Cannes the arrangement was mutually convenient: ByteDance got a flagship demonstration of Seedance 2.0 in long form, and Higgsfield got its recruiting reel.

The reception, for the record, was rough, including from outlets sympathetic to AI tooling. JoBlo’s Chris Bumbray, one of the few critics admitted to the invite-only screening, wrote that the film “screamed ‘AI’ from beginning to end” and left him less convinced about AI features than before. On the supportive side, director Chuck Russell, in Cannes to promote a two-film partnership between his Neumorphic AI shingle and Higgsfield, found the characters genuinely charming while also warning that AI is a wild horse that will run away with a story without discipline, internal logic and a human touch.

For anyone who wants to judge the result directly, the work is streaming on the Higgsfield platform as an episodic series.

Does a two-week, half-million-dollar AI feature read to you as a genuine production shift, or as a tech demo wearing a movie’s clothes? Don’t hesitate to let us know in the comments below!