Arq.live's GLITCH turns voice directions into movies

Filmmaking has always been a painful translation exercise. A director has an image locked in their head, then spends weeks explaining it to cinematographers, set designers, and lighting technicians. Arq.live wants to replace that whole chain of communication with spoken words and a single laptop.

The studio has unveiled GLITCH, a voice-to-film platform designed to bypass text prompts entirely. Instead of tweaking descriptive sentences in a chat box, directors simply speak aloud to build scenes on the fly.

A full short film in eight minutes

Most video generators treat users like prompt engineers playing a slot machine. GLITCH tries to treat you like a filmmaker on set.

Directors speak naturally to cast actors, construct environments, and block scene movement. Behind the interface, a battery of specialized AI models and agents work in sequence, essentially mimicking a digital film crew that renders footage in real time.

During a recent demonstration staged inside a regular hotel room, the team produced an entire short film in under eight minutes off a single laptop, without camera operators, lighting rigs, or traditional gear.

Word travels fast. Michael Shamberg, the legendary executive producer behind Pulp Fiction, Django Unchained, and Erin Brockovich, spotted Arq.live on X and flew in specifically to see it. According to people in the room, the 35-year Hollywood veteran walked out afterward completely speechless.

Built quietly in El Salvador

Surprisingly, GLITCH didn't emerge from a well-funded Bay Area incubator. The team spent the past year building the system in El Salvador, where they previously produced the country's first AI-generated short film.

Arq.live isn't shipping GLITCH as a public consumer toy just yet. The technology is currently available only for serious commercial and narrative productions.

Why it matters

We think GLITCH gets the interface right. Text prompts were always an awkward intermediate step for filmmakers. Spoken direction is the native language of the craft.

By chaining multiple specialized agents rather than relying on a monolithic prompt-to-video model, Arq.live appears to have cracked a much tighter feedback loop. For independent directors who have watched ambitious ideas die in financing meetings, the ability to mock up an entire sequence in minutes from a laptop completely rewrites pre-production economics.

That said, skepticism is warranted. Generating an impressive eight-minute demo in a controlled setting is very different from delivering the shot-to-shot consistency, subtle micro-expressions, and sound synchronization required by theatrical productions. Shamberg's silence could mean pure awe, or it could mean an experienced producer quietly calculating the massive gap between a prototype and release-ready masters. Either way, traditional pre-visualization departments should be paying very close attention.