WorkFlow

Step 1: Create the character sheet first using GPT Image 2

For this one, I made BAOBAO and the senior chef as a fixed stylized 3D reference sheet first. The most important part was defining the characters clearly before animation, so the video generation model could keep their faces, body shapes, outfits, colors, and props consistent throughout the video.

Step 2: Turn the sheet into a 15s cooking comedy video in PixVerse.

The setup: a cozy Chinese restaurant kitchen, one wooden prep table, xiaolongbao, bamboo steamers, warm golden light, flour dust, soft steam, and two characters with very different energy.

BAOBAO works hard.

The master pretends to supervise.

The master secretly eats one dumpling.

BAOBAO catches him.

Then the whole thing turns into a cute martial-arts cooking performance.

I split the prompt by timeline so the model could follow the action clearly:

0-3s dough rolling,

3-5s fast dumpling folding,

5-7s steamer reveal,

7-9s secret dumpling theft,

9-12s steamer-lid blocking choreography,

12-15s warm approval hero shot.

The key was keeping the story simple, the character reference fixed, and the comedy timing very readable.