Ayone

I started by generating one base model image in ChatGPT, then used that same model to generate a separate reference image for each beat the video needed: holding the product, applying a drop, rubbing it in. Each reference image went into Google Flow to generate a short clip, and all the clips were then cut and edited together in CapCut into one continuous ad.

Everything starts with this one image. Since the opening and closing shots are really the same beat, just before and after, I generated both the "checking her face in the mirror" clip and the "checking her face after applying serum" clip from this single base image, so the model stays consistent from the first frame to the last.

The base model image, generated once and reused for both bookend shots.

For the product reveal, I generated a reference image of the model holding the Ayone Serum bottle, matching it against a clean shot of the actual product so the label and bottle shape carried over correctly into the generated clip.

Reference image of the model holding the product.

The actual product shot, used to keep the bottle accurate in the generated clip.

The generated clip: she shows the product to camera, then opens it.

All five clips came together in CapCut, trimmed and sequenced into one continuous ad, mirror check, product reveal, application, rub-in, and the after shot.

The final edited video.