Koh Terai worked as a cinematographer before co-founding a startup. That fact alone wouldn't merit much attention in the crowded world of AI video tools—except that it explains almost everything about what Martini is trying to do.
Most AI video generators ask you to type what you want into a box. Martini, a San Francisco company that emerged from Y Combinator's Winter 2026 batch, built something closer to a virtual soundstage. Position cameras. Adjust lenses. Walk into a generated scene and reshoot from a different angle. The interface assumes you think like a director, not a prompt engineer.
It's a striking bet in a market where text-to-video has become the default paradigm. And it raises a question the film industry hasn't fully answered yet: do professionals actually want to direct AI, or do they just want better results from typing?
Cinematography Meets Silicon Valley
Terai's background straddles two worlds. His work as director of photography on "Grand Bouquet" earned a slot at Cannes Directors' Fortnight in 2019—not the main competition, but a respectable showcase for emerging filmmakers. He also holds a computer science degree from NYU Abu Dhabi and studied design at Stanford's d.school. His co-founder, Long Hoang, handles the technical architecture as "Director of Technology."
The company launched publicly in early February, positioning itself less as a new AI model and more as infrastructure. "Your film set for AI videos," the tagline reads—a phrase that sounds marketing-forward but actually describes the product with unusual precision.
Rather than building yet another proprietary video model, Martini aggregates access to multiple competing systems. Google DeepMind's Veo 3.1. OpenAI's Sora 2 Pro. Kuaishou's Kling 3.0 Pro. ByteDance's Seedance 1.5. Even xAI's Grok Imagine. Each charges per-second through a unified credit system the company calls "Olives"—a branding choice that feels either whimsical or overthought, depending on your tolerance for startup nomenclature.
Pricing spans a wide range, at least as of early this year. Grok Imagine runs $0.06 per second of generated video. Sora 2 Pro costs $0.55 per second. Mid-tier options like Kling 3.0 Pro land at $0.25. The model-agnostic approach means users can test different engines without juggling separate accounts, API keys, and billing systems—an operational headache that anyone who's tried to compare these tools will recognize immediately.
The Interface That Treats Video Like a Set
The core differentiation lives in the controls. Where most platforms foreground a text prompt field, Martini's workspace resembles editing software hybridized with a 3D environment tool. Users can "step into" any generated image or video clip, reposition virtual cameras, and regenerate from the new vantage point. The platform treats AI output not as a fixed result but as a manipulable space.
Start-frame and end-frame controls for compatible models allow something closer to shot planning—establishing visual continuity across generations, a persistent problem when each clip emerges from a separate inference run. For professionals used to storyboarding and shot lists, it's a more familiar mental model than hoping a paragraph of descriptive text captures your vision.
The collaboration layer pushes further into production territory. Martini runs as a web-based editor with real-time multi-user access. Teams can share prompts, review generations in parallel, leave timestamped feedback—"like Figma, but for generative film," according to the company's own description. Projects assemble on a timeline that exports as XML for Adobe Premiere Pro or DaVinci Resolve, hooking AI generation into conventional post-production workflows.
Whether that actually matters depends on your frame of reference. For a solo creator experimenting with AI, the collaboration features might be overkill. For an agency pitching concepts to clients or a production company iterating on pre-vis, the difference between a shared workspace and passing files through Slack could be significant.
The Corporate Structure Gets Murky
Here's where things get slightly odd. The main product lives at martini.film, copyrighted by C47 Inc., which filed for the "MARTINI" trademark in September 2025. But a separate domain, martini.art, branded as "Martini Art," carries a copyright from StoryFlow AI, Inc.—a Delaware corporation registered as a foreign entity in California around the same time.
The relationship between these entities isn't explained on either site. It's possible they're related legal structures for a single company. It's possible they're distinct projects from overlapping teams. Or maybe it's just messy early-stage paperwork that hasn't been tidied up yet. Either way, the dual identity muddies what should be straightforward corporate messaging.
Y Combinator lists Martini's team size as two, though the company's credits page names additional engineering contributors—a common pattern for startups that rely on contractors or advisors without formally expanding headcount. YC's standard deal as of early 2022 was $500,000 total ($125,000 for 7% equity plus a $375,000 uncapped MFN SAFE), though Martini hasn't disclosed whether it accepted those terms or negotiated something different.
The platform currently requires users to request access via Typeform—early access friction that's typical for resource-constrained startups but also signals the product isn't quite ready for open availability. Four pricing tiers range from a free "Indie" plan (public projects only) to enterprise contracts with custom SLAs. The Pro tier costs $30 monthly and includes $30 in generation credits; Studio runs $150 with team features and $150 in credits.
Context and Competition
Martini enters a field that's simultaneously nascent and already crowded—a situation that tends to produce both opportunity and chaos. Runway has been building toward professional adoption with its Gen-4.5 model and world simulation tools. Adobe folded AI features directly into Premiere Pro and launched a standalone Firefly video editor. Lightricks announced LTX Studio. Moonvalley released its Marey model specifically for licensed commercial production last summer.
Each takes a different stance on the control-versus-simplicity spectrum. Some prioritize ease of use for non-experts; others optimize for integration with existing creative software. Martini's director-focused interface carves out a distinct position, though whether that resonates with actual production studios remains an open question.
Former Adobe executive John Nack highlighted Martini's "stepping into the set" capability in a February blog post, noting the closed beta status at the time. VP-Land's analysis that same month positioned the platform as a "film set mentality" layer atop raw AI models—different from single-model generators and more control-oriented than Google's Flow tool, which the company previewed at I/O last year but hasn't fully released.
The company claims more than 200 films have been made using Martini, "aired as TV commercials, displayed as artwork, and gone viral." Those figures are self-reported and unverified by outside sources, the typical situation for early-stage product metrics. The showcase site, Made With Martini, features work by creators including Elad Segoli, Gami Orbegoso, and Andy Oleck—names that won't be familiar to most people but carry some weight in adjacent creative communities.
A Bet on Professional Workflow
One detail worth noting: Martini lists ByteDance's Seedance 1.5 among its integrated models, though ByteDance released Seedance 2.0 in China in February. The newer version faced delayed global API access and pushback from Hollywood industry groups concerned about Chinese data practices, leaving its status for Western markets unclear. Whether Martini eventually integrates 2.0 or sticks with 1.5 probably depends on regulatory and business considerations well beyond the startup's control.
The broader bet Martini is making centers on a simple premise: that professionals want to direct AI rather than merely prompt it. That video generation should feel less like asking a genie for a wish and more like working with a crew on set—even if the set is virtual and the crew is algorithmic.
It's too early to say whether production studios, agencies, and independent filmmakers will embrace that metaphor. Some directors might prefer to offload creative decisions entirely, using AI as a fast-drafting tool. Others might find Martini's approach captures something closer to their existing mental models.
But in a space where most tools still put the text box front and center, treating the interface like a film set rather than a search bar is at least a notably different philosophy. Whether it's the right philosophy will depend on how the next generation of filmmakers—raised on both Premiere Pro and generative models—actually wants to work.
