Google DeepMind's Veo joins Hollywood for Tribeca short film

Filming a newborn baby under harsh studio lights is a logistical nightmare for any director, while building one with traditional CGI almost always trips straight into the uncanny valley.

For her new short film ANCESTRA, director Eliza McNitt found a third path. She directed live actors holding essentially nothing, and then let Google DeepMind's video model, Veo, generate a realistic newborn right into the live-action frame.

The project, which just premiered at the Tribeca Festival, is the first of three shorts born out of a partnership between DeepMind and Primordial Soup, the storytelling venture founded by director Darren Aronofsky. Over 200 traditional film crew members, VFX artists, and sound designers worked alongside DeepMind researchers to make it happen.

Teaching an AI to follow camera rigs

Generative video usually feels like a slot machine where you pull a prompt lever and hope for the best. Film directors cannot work that way.

To lock down the emotional look of the film, which was inspired by McNitt's own dramatic birth with a heart defect, the team fed actual vintage photos taken by her father into Gemini. The LLM broke them down into hyper-specific aesthetic descriptions, which Imagen then turned into consistent concept art.

The real breakthrough was controlling the camera motion. In one ambitious sequence, the camera travels through a human body and into the womb to depict a C-section birth.

Instead of praying Veo would guess the right pan and tilt, artists built a rough 3D human body and keyframed a draft camera move. Veo was then instructed to track that exact movement and render the final scene with the same motion. Another sequence showing organic tissue closing up took minutes to generate using reference video matching, skipping days of traditional 3D simulation.

Blending generative pixels into classic pipelines

Rather than replacing Hollywood, Veo acted more like a specialized VFX brush inside a standard production workflow.

For the birth scene, McNitt filmed real actors, masked out the target area in the shot, and used Veo's new object-addition capability to insert the AI-generated baby. The shot was then passed to traditional VFX artists and colorists to match film grain, lighting, and motion blur.

Another complex shot, peering out from inside a cracking crocodile egg at sunset with the mother nearby, layered several Veo and Imagen assets directly inside a traditional compositing suite.

Why it matters

We think this project reveals the realistic future of AI in cinema: not as a magical one-click replacement for film crews, but as a high-powered puzzle piece slotted into existing pipelines.

Pure AI films often feel like floating slideshows because text prompts lack spatial precision. By introducing camera-motion matching and live-action object insertion, DeepMind solved the control problem that has kept serious filmmakers away from generative models.

For indie filmmakers, this hybrid approach opens huge doors. Visualizing internal anatomy or complex biological textures usually requires high-end CGI budgets. Being able to sketch a 3D path and let an AI handle the photorealistic rendering levels the playing field significantly.

Still, it is worth keeping in mind that this was a high-touch showcase with Google's engineers tweaking models behind the scenes. Until these precise motion-matching and inpainting tools become standard in commercial software, most creators will still be stuck wrestling with prompt randomness.