generating ai video, locally
If you have the hardware, you too can waste many hours creating 6 second videos of dogs wearing funny hats and eating fruit.
the tools
We are using comfyUi and the Wan 2.2 models. You can do this online, sure, but why would you?
All you need is 96 gigabytes of RAM and a graphics card equivalent to a couple of car payments.
And another half-terabyte of disk space to hold the models.
start with a source image
If you have a specific idea in mind, start with a source image.
“I want fruit that looks like it was made out of stryofoam”.
In my case, I found a stock image of a strawberry, and then used image-to-image to refine.
generate some video
When generating video, get used to waiting. Batching overnight is the only way to go.
On average, it is going to take 5-20 minutes to generate 6 seconds of video.
where i started
i don’t hate it
so much liquid
such texture and realism
in conclusion
Look, if you do this you are going to waste a whole lot of time and electricity. And for what?
If you need to ask yourselves these questions, it isn’t for you. You are looking at hundreds of iterations over the course of several weeks to get to the final result.
Checking the batch was like Christmas morning.