CureBay built its Kavach ad scene by scene in Gemini and Veo 3.1

CureBay's Kavach conversation needed a believable village setting — a Swasthya Mitra talking to a group under a banyan tree — with several people, positions and outfits staying consistent across scenes.

Build the ad as a chain of 10-second Veo scenes anchored to one Gemini Pro reference image — the group under a banyan tree is the continuity system.

Scene stills first in Gemini Pro (people, seating, clothing, tree, light), then 10-second Veo 3.1 generations per dialogue beat — introduction, product, ₹499 annual benefit, a villager's question — assembled in edit.

The finished film holds a multi-person group conversation with consistent characters across cuts — a workflow the studio documented, including the voice-generation fight with the word 'loan'.

For the CureBay Kavach film, production studio Portray Media (Manas Ranjan Behera) built the ad scene by scene rather than betting on one cinematic prompt. The setting is deliberately communal: a Swasthya Mitra — CureBay's community health worker — talking with villagers under a large banyan tree, so the message plays as a local conversation rather than a staged studio advertisement.

Stage one established the visual anchor as a still in Google Gemini Pro: the Swasthya Mitra, the group of villagers, seating positions, clothing, background, lighting and camera composition. 'The group is not background decoration. It is part of the continuity system' — the prompts kept insisting on preserving the uploaded image, so nobody changed seats, vanished or changed clothes between scenes.

Stage two divided the conversation into scenes of at most 10 seconds each in Google Veo 3.1, cut on natural dialogue beats: the introduction, CureBay Kavach itself, the ₹499 annual benefit, a villager's curious question about how the offer works. Each generation carried one clear communication purpose, and the scenes were assembled during editing.

The write-up is candid about the hard parts, including voice generation stumbling over the word 'loan'. The result is a multi-character, continuity-heavy village ad produced end to end with Gemini and Veo — and a documented workflow for doing it again.

A group scene multiplies the continuity problem — several people, positions and outfits to hold constant across generations.

The 10-second generation window forced one clear purpose per scene, which kept prompts small and failures cheap.

The Gemini reference image did the world-building once, so Veo only had to handle motion and dialogue.

Dialogue beats gave the edit natural seams, so the generation limits never showed in the finished film.

With generative video, continuity is a system: anchor everything to one reference image and let each generation do exactly one job.

Portray Media published the full workflow — reference image, 10-second scenes, dialogue, performance, final prompts, edit — as a repeatable playbook for AI-produced ads; the article is dated 11 October 2026.