Suvasis Mukherjee

The bay has swallowed the streets. Tireless service androids do all the labor, and the humans they tend have grown languid and dependent — cared for, and quietly obsolete. The film moves through three acts: the San Francisco we still recognize, standing in the fog; the deluge, with a drift through drowned mangroves as the threshold into a changed world; and the machine-tended new world that rises from the water — arcologies, hover-skiffs, and androids carrying on while people recline. Every frame was generated during the hackathon on fal — FLUX Pro for stills, Kling v3 for image-to-video, ElevenLabs for score, MiniMax for narration — assembled by a config-driven pipeline. Its core is an autonomous judge-in-the-loop: a Gemini judge scores each image and clip against a calibrated rubric; strong shots auto-accept, weak ones trigger automatic prompt-revision and regeneration, and only borderline results escalate to a human. AI generating, evaluating, and correcting AI — the human left only at the ends, where taste actually matters. code is in https://github.com/dronomyio/fal_ai_hackathon.git, and the session url is https://fal.ai/dashboard/recent-history.