The coffee ai video prompt
The coffee ai video prompt has to do three things at once: copy the motion of a reference clip, keep two different people looking like their photos, and put every spill on the right second. On Coffee AI Video that is already built in, so there is nothing to type. Make the video from your photos.
What the prompt has to solve
- The reference clip. The skit is a precise piece of movement, with a spoon, a kettle and a jug. Words alone do not describe it, so the model needs the original clip to follow.
- Two identities. Two people are in the shot, each one from a different photo. The model has to keep them apart, and the person standing further back is the hardest to keep.
- The timing. The first miss lands around 2 seconds, and the milk tips at about 12. Move that around and the joke stops working.
Why copied prompts fail
People post about the same few failures. The first is both faces turning into one face, because a pasted prompt has no way of saying which photo is which person. The second is the milk arriving early or never, because the timing lived in the reference clip and not in the text. The third is a mug that fills itself or a kettle that pours into nothing, because the model fills a gap with what is most common. The fourth is the wrong roles, with the person who should pour holding the spoon.
Each failure costs a paid attempt, and the fix is rarely another sentence in the prompt. If you want the cost side, read how to make the video, which compares the built-in route with doing it yourself.
What is built in here
The motion reference, the prompt for each template and the settings are fixed, so every render follows the same shot list. You choose who is in the shot and which kitchen. The video model is Seedance 2.0 / 2.5, made by ByteDance, and you pick the tier in the generator. Results can still vary, as faces and hands drift a little, but the order of the beats stays the same. To pick a role, browse the templates.