How a $2,000 AI-generated feature reached Tribeca — and what it means for film and business

TL;DR / Key takeaways

• Dreams of Violets is a 75-minute drama created by Ash Koosha with generative AI tools and is screening at the Tribeca Film Festival; Koosha says the project cost under $2,000 and took roughly 2.5 months.
• Every on-screen image and character was AI-generated; the script was written by Koosha with structural help from the chatbot Claude, and Koosha voice-acted all roles before applying AI voice-modification.
• Generative AI can dramatically lower costs and speed production, creating new business opportunities and job types while raising legal, ethical, and aesthetic questions—particularly for politically sensitive subjects.
• Media leaders should pilot AI workflows, update IP and licensing contracts, and adopt human-in-the-loop ethics protocols before scaling AI production.

Content note: Dreams of Violets draws on eyewitness accounts of recent protests in Iran. Some figures and reports about casualties are contested; the film avoids using the likenesses of living individuals in Iran to reduce safety risks.

AI-generated film at Tribeca: what happened and why it matters

A 75-minute film made for under $2,000 is screening at Tribeca—and that single fact reframes how producers, studios, and executives must think about cost, speed, and creative control. Dreams of Violets, directed by Iranian-British artist and entrepreneur Ash Koosha, is being billed as a fully AI live-action feature accepted at a major festival. Every visual and on-screen character was produced with generative AI; Koosha wrote the script, refined structure with Claude, voice-acted every role, and used AI to alter the vocal characteristics.

“I spent under $2,000.”

Koosha describes his intent plainly: “I’m not selling AI. I’m just trying to use a tool to tell a story.” Yet the project does more than prove a technical trick—it exposes business implications that deserve strategic attention.

How it was made: the practical pipeline

Quick summary of the workflow:

• Script & structure: Koosha wrote the screenplay and used the chatbot Claude to iterate language and story beats.
• Visuals & characters: Generative image and video models created backgrounds, faces, and on-screen action.
• Performance layer: Koosha voice-acted all parts and then applied AI voice-modification; score and editing were handled manually (no AI).
• Safety choices: Characters were not modeled on living Iranians to reduce real-world risk to subjects.

Koosha frames the creative loop like this:

“You just open another session. You don’t have to worry that you’re rewriting. You multiply your imagination until something hits the right spot.”

That iterative, session-based workflow is a core advantage of generative models: rapid prototyping without expensive physical shoots, location permits, or large crews. Contrast a traditional VFX-heavy sequence—which can cost hundreds of thousands to millions for complex shots—against a generative pass that can be produced in days with modest compute and software credits. For indie teams, that difference can be existential.

Economic implications for studios, vendors, and indie creators

Two economic shifts matter most.

1. Unit-cost compression. For many visual needs, the marginal cost of producing images and scenes can drop dramatically. That doesn’t erase high-end cinematography budgets entirely—some tentpoles and practical-effect-driven movies will still justify large budgets—but it forces a reassessment of where studios invest. Koosha predicts seismic change:

“Well, I don’t think Christopher Nolan will make another $300m movie. Underwriting a $200m to $300m movie will not make sense any more.”

That’s a bold projection and likely exaggerated in the short term. Big-budget filmmakers still sell global star power, theater spectacle, and practical stunts that models can’t wholly replace. But for mid-budget visual-features and rapid topical storytelling, AI lowers the bar to entry and reshapes cost-benefit math.

2. New job categories and business models. Koosha expects Fountain 0 and related ventures to create roles that didn’t exist a few years ago:

• AI narrative director / creative technologist — combines storytelling with model orchestration.
• Prompt designer / prompt engineer — crafts inputs to produce nuanced emotional outputs from models.
• Model license & provenance manager — tracks model training sources and legal exposure.
• Ethics reviewer / trauma sensitivity officer — ensures depiction of real events respects subjects and safety.
• Voice & likeness compliance officer — manages consent, payments, and contracts for voice/face use.

“I guarantee that this company will create at least 200 jobs that didn’t exist.”

That optimism is plausible—jobs will appear, but they’ll differ from traditional VFX and acting roles. Some existing tasks may be displaced, others will be elevated toward oversight, curation, and rights management.

Ethical, legal and safety risks

Using generative AI to depict recent political protests amplifies responsibility. Even with anonymized AI-generated faces, producing realistic scenes of violence or trauma risks misinforming audiences, exposing real people, or being weaponized for propaganda. High-impact claims should be handled carefully: Koosha says “80%” of the film is a recreation of events based on eyewitness accounts—those are subjective reconstructions, not documentary verification.

“I would say 80% of it is a recreation of events that actually happened.”

Legal frameworks and commercial contracts are still catching up. A few concrete exposures and recommended contract clauses for rights teams:

• Provenance clause — require documentation of model training data sources and licensing to defend against IP claims.
• Consent and compensation clause — specify how likeness or voice models derived from live performers are licensed and paid.
• Indemnity for deepfake misuse — allocate financial responsibility and disaster protocols if generated content is repurposed maliciously.
• Security red-lines — forbid generation of content that imitates currently living private persons without documented consent, especially in politically sensitive contexts.

Unions and guilds have already voiced concern about AI’s effects on performers and crew; organizations like SAG‑AFTRA are actively negotiating protections around voice and likeness. Regulators may follow with provenance and labeling requirements—festivals and platforms will likely ask creators to disclose model usage and sourcing.

Artistic quality: augmentation versus replacement

Koosha is candid about his ambivalence toward current AI aesthetics:

“So far, I hate anything made that is made with AI. It disgusts me. I don’t want to look at it. It gives me a headache.”

His solution is restraint: use AI to augment craft, not to replace creative judgment. That hybrid model—AI for imagery and rapid ideation plus human narrative control, sound mixing, and editorial judgment—may be the most sustainable route. For emotionally complex performances, human actors and directors still bring nuance that models struggle to replicate consistently.

What business leaders should do next

For executives in media, advertising, and studios, a pragmatic playbook avoids panicked bans and naive adoption. Actionable steps:

• Run a three-month pilot: Use generative AI for one production phase (previsualization, background generation, or concept art). Track time-to-proof, cost per minute of screen, and audience response.
• Update contracts: Add model provenance, consent, and indemnity clauses to procurement templates.
• Upskill staff: Train a small cohort in prompt design, model evaluation, and human-in-the-loop oversight.
• Establish an ethics gate: A multidisciplinary review for politically sensitive content—legal, editorial, and trauma experts must sign off.
• Measure KPIs: Proof-of-concept time, cost delta versus traditional methods, legal incidents, and audience trust/reception.

Quick executive checklist

• Assess IP & licensing exposure within 30 days.
• Budget a small pilot (under $50k) to validate workflows.
• Create a model-provenance log for every AI asset used.
• Assign a named ethics reviewer to all politically sensitive projects.
• Negotiate model-usage warranties with vendors.

Ethics in practice: a mini-framework

Operationalize responsible AI filmmaking with four principles:

• Human-in-the-loop: Every final creative decision should involve a named human accountable for artistic and ethical outcomes.
• Source transparency: Document what models and datasets were used and why.
• Consent-first: Default to not using or imitating living persons without clear consent, especially in sensitive contexts.
• Trauma sensitivity: When representing recent violence, work with subject-matter experts and limit sensationalized imagery.

Counterpoints and where optimism meets friction

There are credible reasons to be skeptical. Festival programmers and critics may treat AI-made films as novelties until the craft matches human-led storytelling consistently. VFX houses could see pressure on mid-tier gigs, and unions will push for protections. Legal standards—on who owns an AI-created face or how training data was sourced—are unsettled and ripe for litigation.

On the other hand, generative AI democratizes visual storytelling: smaller teams can tell topical stories quickly and at low cost, and new businesses can form around creative-technical services. The question for executives is not whether AI will be used—that is already settled—but how to shape its adoption so it creates value, preserves trust, and protects people.

Final thought for decision-makers

Dreams of Violets is a proof point: generative AI for film production is practical, fast, and cheap enough to matter. It will not erase all human craft, nor will it be a perfect substitute for actors and editors. But it forces a strategic choice—ignore the change and cede ground to nimble competitors, or build governance, contracts, and capabilities now so AI becomes an engine for new creative and commercial models rather than a source of risk.

“I’m not selling AI. I’m just trying to use a tool to tell a story.”

That modest line contains a big prompt for leaders: treat generative AI the way you treat any transformative production tool—test it judiciously, regulate its use responsibly, and design new roles and contracts to capture the upside while limiting harm.