Skip to main content
Live
Main content

Tribeca 2026 shows AI filmmaking needs custom models, not vanilla prompts

Google DeepMind and OpenAI brought AI shorts to Tribeca. The standout used custom Veo and Imagen builds trained on hand-painted Pixar concept art.

Jaeden Schafer
Editor in Chief · · 5 min read
Google logo

The most polished AI film at this year's Tribeca Film Festival was not built by typing prompts into a stock video model. Google DeepMind's Dear Upstairs Neighbors, directed by Pixar veteran Connie Qin He, ran on custom versions of Gemini-adjacent models Veo and Imagen that were fine-tuned on concept art hand-painted by Pixar production designer Yingzong Xin. The result was the clearest signal yet that commercial AI filmmaking will route through bespoke pipelines, not consumer-grade text-to-video.

Tribeca, which ran shorts from Illuminai Studios, Asteria Film Co., DeepMind, and OpenAI on June 13, 2026, became a side-by-side test of two approaches. The vanilla-model entries — Illuminai's Roar and Asteria's ChikaBOOM! — read as disjointed montages of inconsistent clips. The customized pipelines held together as cinema.

Most AI firms' video models are still only capable of churning out short bursts of visually inconsistent footage.
Charles Pulliam-Moore, Film & TV Reporter

Dear Upstairs Neighbors follows Ada, played and produced by Márcia Mayer, as she tries to sleep through the noise of her upstairs neighbors. The painterly look He wanted is the kind of style that off-the-shelf video models typically smear across frames. DeepMind's engineers built versions of Veo and Imagen tuned to Xin's concept art so that shots would stay consistent with the project's visual language.

Key facts

  • 01Google DeepMind's Dear Upstairs Neighbors used custom Veo and Imagen builds trained on Pixar designer Yingzong Xin's hand-painted concept art.
  • 02OpenAI's Critterz lost its Cannes Film Festival slot after the company shut Sora down entirely.
  • 03Ash Koosha produced Dreams of Violets solo for $2,000 in compute using Kling AI, Claude, Gemini, and Nano Banana.
  • 04Tribeca 2026 ran AI-forward shorts from Illuminai Studios, Asteria Film Co., Google DeepMind, and OpenAI on June 13, 2026.
  • 05Dear Upstairs Neighbors built scenes in Autodesk Maya first, then passed roughs into Veo for stylized polish.

The workflow was not pure generation. Production used Autodesk Maya — the standard for 3D rigging and VFX — to block rough animations first. Those roughs were then fed into Veo to be polished and dressed with stylized assets generated by Veo and Imagen. The model handled finish; humans handled blocking, staging, and creative direction.

OpenAI's contributions told a different story. Alice Gu's semi-autobiographical Smoked used Sora to recreate the Palisades Fire, with wide shots looking cartoony and close-ups filmed on a Volume-style setup faring better. Youssef Michraf's Mauvais Soleil leans on photorealistic Sora-generated shots that mostly last a few seconds, with a single unseen narrator — choices that double as workarounds for the model's limits.

OpenAI's presence at Tribeca was notable given that the company recently shut Sora down entirely. That shutdown is why OpenAI's feature-length film Critterz could not debut at this year's Cannes Film Festival. The company appears to be pulling back from video-focused product bets, even as its tools remain in active use among independent filmmakers.

It's important to bear in mind that Dear Upstairs Neighbors would not be nearly as visually impressive if it had been produced with vanilla versions of Google's various models.
Charles Pulliam-Moore, Film & TV Reporter

The economics of the indie tier are worth pausing on. Writer and director Ash Koosha produced Dreams of Violets, a docudrama about the protests in Iran over the past year, by himself for $2,000 spent entirely on compute. He stitched together Kling AI, Claude, Gemini, and Nano Banana over a few weeks. The narrative carries the film; the visuals do not break new ground.

That gap — between a $2,000 solo docudrama and a DeepMind-engineered short with a Pixar director — is the real shape of the market. Vanilla generative tools have collapsed the floor on what a single person can produce. Custom-trained models, attached to traditional animation pipelines and named studio talent, are where anything resembling theatrical quality is happening.

Related · from this week
Nobel economist Daron Acemoglu names three AI shifts he's watching now
Jaeden Schafer · 5 min read →

Skeptics inside the festival circuit will note that Dear Upstairs Neighbors is also, functionally, a commercial for Google's technology. The custom Veo and Imagen builds were developed in collaboration with DeepMind researchers, and the workflow's success says as much about DeepMind's willingness to staff a film as it does about the models themselves. Roar and ChikaBOOM! suggest that without that kind of in-house engineering support, the off-the-shelf tier still produces work that struggles to hold a viewer's attention.

The competitive read for the AI labs is straightforward. The studios that matter are not going to license a generic API and ship features. They are going to demand fine-tuned models trained on their own art, integrated with Maya and the rest of the existing VFX stack, and operated by their own artists. That is a services business as much as a software business, and it favors labs willing to embed engineers on productions — which is exactly the pattern DeepMind ran at Tribeca and the pattern OpenAI just stepped back from when it pulled Sora. Whoever staffs up to do bespoke model work for studios over the next twelve months will define what the next decade of AI-assisted film actually looks like.

ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from Analysis

Nobel economist Daron Acemoglu names three AI shifts he's watching now
Analysis

Nobel economist Daron Acemoglu names three AI shifts he's watching now

The 2024 economics laureate still doubts an AI jobs apocalypse, but agents, in-house economists, and missing apps are on his radar.

Jaeden Schafer5 min read
AI's underpants gnomes problem: Mercor study finds agents fail most of 480 workplace tasks
Analysis

AI's underpants gnomes problem: Mercor study finds agents fail most of 480 workplace tasks

A February 2026 Mercor study tested top agents from OpenAI, Anthropic, and Google DeepMind on real banker, consultant, and lawyer work. They flunked.

Jaeden Schafer5 min read
Google logo
Models

Google ships Nano Banana 2 Lite and Gemini Omni Flash to developers

DeepMind's fastest image model generates in 4 seconds at $0.034 per 1K images; Omni Flash matches Veo 3.1 Fast at $0.10 per second of video.

Jaeden Schafer5 min read