Fridge drawer tiramisu, with two suspects.
Two familiar faces and one shared midnight mission. Our Duo demo brings two illustrated characters into the same inside-fridge scene, with a single tiramisu tray connecting the action.
- 10 seconds
- 9:16 vertical
- AI-generated fiction
Demo available now. Public generation and purchases are not open yet.
Preview duo mode ↗Watch without an account. Video model: Kling 2.5 Turbo.
Published November 2, 2026 · Curated for pacing
A shared tray makes a two-person story
The “fridge drawer tiramisu” setup here is a dessert tray on a shelf, viewed from inside the refrigerator. Duo adds a second character to the same compact space. That changes the scene from a private snack to a shared secret: the viewer needs to understand who is reaching for the dessert and how the other character responds.
Both faces need room in the frame. The creative direction keeps the tray close to the camera and the characters beyond it, leaving their expressions readable instead of filling the shot with overlapping arms. The point of view stays fixed so the relationship between the two subjects remains the center of the joke.
Watch the turn-taking and empty-tray ending
In a Duo clip, the interesting detail is the exchange between the characters. Follow the reaching and eating action, then look at the shared reaction once the tray is empty. Our target is to complete the snack around six seconds and use the final seconds for that look. This published demo was lightly retimed after generation.
Two subjects create more opportunities for visual mistakes. Hands can overlap, a face can drift, and the model may make the action less coordinated than intended. The demo is a selected example, not a guarantee of precise synchronization or identity preservation for every pair of photos.
Why Duo starts with two separate references
Prepare one clear image per person or character. Separate portraits make it easier to identify the intended pair than a crowded group photo. Similar image clarity and a compatible visual style can make the combination easier to judge. Avoid a heavily filtered close-up for one subject and a tiny, blurred image for the other.
Only use material you own or have permission to use, including permission from both people if you plan a real-person scene. Public real-person generation is currently closed. Our demo uses the original illustrated young man and woman shown below; the shared fridge-scene frames were prepared separately before Kling 2.5 Turbo animated them.
How this demo was made
The video uses Kling 2.5 Turbo (Image to Video Pro) through Kie.ai. Its opening and ending scene images were prepared separately with OpenAI image generation. We do not have a verified model-version identifier for those demo frames. They are curated examples of the direction, not unedited customer outputs.
The configured customer image model is Nano Banana 2 (Google Gemini 3.1 Flash Image), also accessed through Kie.ai. Learn more in our model disclosure, or read the fridge cam video guide for a closer look at the viewpoint, pacing and AI limitations.
