I build

Ground the character in something real. Canela began as photos of my actual cat — name included; Ciby began as Lider's existing mascot from past campaigns. Real references give AI characters a soul generic prompts can't.

Give each character a job in the store world — Canela works checkout and restocks aisles, Ciby runs missions. Scenes are written as story beats, so every video is a small episode rather than an ad.

Lock "hero frames" with the client before generating any motion — canonical stills that define look, wardrobe, lighting, and brand fit. Sign-off happens on frames, which are cheap to iterate, not on videos, which aren't.

Let the audience steer. Canela was meant to be a one-off generic cat — client and audience response turned her into a recurring character.

Curate reference sets — real photos for Canela, past campaign frames for Ciby — covering angles, lighting, and expressions the character will need.

Generate hero frames with NanoBanana, iterating until identity holds across contexts: same face, same wardrobe, same brand palette in a checkout, an aisle, or a spaceship.

Drive video generation with Google's Omni model from those frames. Because shots are generated from shared references, changing a character or prop at any point propagates through previous and upcoming shots — no regenerating from scratch.

Finish like a film: Topaz upscale to broadcast quality, voice and sound with ElevenLabs, then subtitles, motion graphics, and audio mix in post. Delivered as 16:9 film and 9:16 social cuts.