A fish kept alive by a dripping tap took second at Venice's AI festival (interview)

Alex Naghavi, a creative director of 17 years who led the 2012 Myspace redesign, was making her first AI films; the logline came from a screenwriter friend in Australia.

An AI short assembled like a production system: every shot gets its own node chain in Flora, so all 70 could be re-pointed to a new video model overnight.

Ezra Li voiced the fish himself — hours of French takes in a parked car, morphed with ElevenLabs, no generated voiceover — while Nano Banana, Claude and Seedance ran inside per-shot Flora workflows, with live Foley replacing generative sound.

Le Drip took second place at the Reply AI Film Festival in Venice on 6 September 2026, out of more than 3,000 shorts from 76 countries.

In a creator interview with The Daring Creatives, Alex Naghavi and Ezra Li describe how Le Drip, their painterly AI short about a fish in a kitchen sink, took second place at the Reply AI Film Festival in Venice on 6 September 2026, out of more than 3,000 shorts from 76 countries. The film is in French: the fish argues with the dripping tap that keeps it alive, about God and dying, while a harbour town carries on outside.

Screenwriter Jessie Hughes supplied the one-line premise. Naghavi, a Josephmark partner who led the 2012 Myspace redesign, says she stalled on a silent animatic until Li improvised the fish's inner monologue as a Frenchman and then wrote the screenplay. Li refused a generated voiceover: he recorded hours of takes in a friend's car in a London parking lot, in a French he says he doesn't really speak, then had ElevenLabs morph his own intonations.

Most of the film was built in Flora, a node-based canvas running Nano Banana for images, Claude for prompting and Seedance for video, with every shot wired as its own sequence. With the film 90 percent done and rendered in Kling, Li suggested testing Seedance 2.0 the day after its US launch; Naghavi says it held frames and got the painterly, stop-motion movement they were after, so they redid all 70 shots in one weekend.

Sound designer Laurent Delforge found generative sound too loose for the close ASMR feel of a fish in a sink and recorded live Foley with Olivier Thys; colorist James Foundoulis graded the shifts between shots. The credits also list Runway for storyboarding, Adobe for video and sound, and a custom end-credits tool built in Google AI Studio; both makers are Runway Creative Partners who met in the first cohort of Google's Flow Sessions residency, started in September 2025.

Naghavi ran the AI film like a design system: per-shot workflows kept prompts attached to shots, so switching video models meant re-pointing nodes, not hunting through chat histories.

Human craft stayed in on purpose — a recorded, morphed voice and live Foley wherever generation felt wrong.

A two-person team borrowed film roles where they mattered: a screenwriter for the premise, sound designers for texture, a colorist for consistency.

Treat generative shots as a database, not a sequence: a per-shot prompt structure lets a two-person team re-render an entire film over one weekend when a better model lands.

The Venice podium came roughly 18 months after Naghavi's first AI films. Li's earlier short HOME had already won the Grand Prix at Runway's Gen:48 Fourth Edition. Naghavi says next time she would bring the colorist in during image generation to hold consistency from the start.