It's been a while since I had anything to post, but I've had good results with Grok latest version.

It needs a lot of babysitting and good image files (this is entirely I2V).

Hope you'll like it and let me know if you need prompts or tips.

Also one guy gets pied at the end, just FYI

https://youtu.be/DsoGiYR8rWM

Loving the two mean teasing secretaries at the end! That's my kind of office job..

Wow. Thanks for sharing. Can you please share any prompt examples if you don't mind. I love fooling around with Grok, and haven't figured out how to get good pie prompts yet.

Seconded!! Thanks for sharing

So here is a bit more info and tutorial as well as examples. First some basic rules about Grok imagine:

a) use 6 seconds mode, not 10 seconds. They seem to have the same "token budget" so 10 seconds will drift vastly more. Always test in 6, then you can switch to 10 and see if it works. Also use 480p and then the upscale function when you're happy with the result
b) always use image to video, as the tool will really perform badly otherwise;
c) use the highest quality source image you can get, since it will be the ceiling for the quality of the video. Try to use images with simple, easy to understand perspectives without complex angles or occlusion.
d) use short prompts, maximum 1-2 paragraphs as the tool will drift quickly with longer prompts
e) although the tool can technically do dialogs, I wouldn't recommend it as it works much better with the default music
f) don't bother specifying lighting, camera, etc. It won't work for I2V
g) try to stick to a single sequence, and avoid characters leaving and coming back (they will change)
h) avoid complex backgrounds or crowds, everything that needs to move has a cost
i) try to use a source image with distinct characters, eg. different outfit colors, hair colors, etc. Don't use names as it doesn't work well.

Then the process is simple. You want to make it easy for the system to render the pie, so you rotate the camera for a side view. You decompose the movement so that there is no ambiguity on the position. You force the cream to not be liquid or splattery. And that's it really.

Here is an example, source image first and key frames after


Rotate the camera over the scene until the blue girl is seen from the side

blue girl is acting embarrassed and nervous for losing the game.

Maid holding the pie takes her time, perfectly positioning the frosting cream pie paper plate for the perfect pie in the face movement. Now very close, she open palm shoves and smears the pie forward (fast and brutal motion), driving the pie squarely onto the woman's face with comical slapstick force, shoving her back. Thick, stable, sticky frosting smooshes and adheres heavily, covering her face completely with minimal splatter--mostly slow-falling chunks and gradual drips. The woman recoils, with no expression visible under the frosting cream. Dirty creamy plate is pulled back after the pieing by the girl who brought the pie.

Camera then zooms in on the pied girl's face while a Japanese tv overlay appears with bright colors




It's much simpler to use than SORA, and really just a matter of telling the model what to do.

If you have any problem let me know!

Thank you so much for sharing. Now I know what i'm doing for the rest of the day.

lol ideal Valentine's Day plan

This is very impressive. Getting pie sandwiches and convincing motion has been difficult up to now.

Been playing with this some ... results attached, fairly similar to each other. But fun I think.

The starter image is important, definitely have a better time when the pie is already in the right position.