Task:fill product cards with additional photos at catalogue scale — in the admin and on the public side — with batch food photography pipelines and img2video.
From one SKU to a finished cardthe photos and the brand rules go into the database; the database and the prompt go to the AI director, who hands each frame to a model; every frame passes the regression tests before it takes its slot in the card — and the card goes from the admin to the storefront
SKU photosseveral shots of one product, side by sideAdmin → Brand settingsthe rules, formalised and entered by handDatabasephotos and rules together, straight to the directorPromptan input of its own, alongside the dataAI directorholds the character and the locations; picks the modelSeedreamimage modelNano Banana 2image modelSoravideo modelKlingvideo modelControlunit tests against quality regressionProduct cardextra photos, slot by slotStorefrontfrom the admin to the public side
How it works
Formalised and entered in the admin; the prompt is built from them and the SKU photos
Holds the character and the locations across the batch, picks a model per frame
15+ iterations with the character and the locations pinned
Unit tests against quality regression
The extra photos land in the card and go from the admin to the storefront
Results
15+
prompt iterations kept
4
model families in the pipeline
2
retail chains
Catalogue-scale imagery
Card and campaign material without booking a studio for every SKU
Quality does not drift
Regression is caught by tests rather than by the client noticing
Stack
Seedream · Nano Banana 2 · Sora · Kling
Something here look like your problem?
Tell us what you are building. A paragraph is enough to start.
and 