Video AI tutorial

Disco Diffusion v5.2

Go here: https://colab.research.google.com/drive/1q_g0ntxagI77t4mHdNFKHYk4ersiNsQg?usp=sharing

**1 – This is how Disco Diffusion looks like**

**2 – First things first! Save your copy**

`File - Save a copy in Drive`

It will create a copy in your GD account

3 – **Go max with hardware and RAM** if possible

Notebook Settings - GPU - High RAM

Since we are using Google’s GPU, to save some time, and energy, and just really, simply, respect your nervous :), I would recommend getting the **Collab Pro** plan, whichever is affordable for you. But you can still do it without any of their plans

Connect - (then choose whatever fits you, since I have Collab Pro, I am going to go with the hosted runtime)

it should look like this when connected
if you have the **green checkmark**, you’re good to go!

5 – Choose (your destiny!!!) the video that you want to F up

I’ll go with this one! Can I? Thank you.

The original file resolution is too big, so I downsized it, made it smaller
Bigger res you have, more time it will take for AI to process each frame, so! Up to you.

Now resolution I have for this video is **1622×640** px

6 – Upload your video

Regarding the settings and all that fun/complicated stuff

I’m going to skip, and just drop the screenshots, for you to see the most important setting that I used

Those numbers are very important, and there are so many good tutorials out there, better than me explaining how all of it works. As an option, AI itself has tons of links/docs inside, feel free to invest some time into readings. It will help you better “play” with final outcomes

or, just simply copy-paste these numbers until you figure it out)

**batch_name:** YOUR PROJECT NAME GOES HERE
**width_height:** PUT YOUR VIDEO RESOLUTION OVA HE-AA

in **Video Input Settings:**
**video_init_path:** MAKE SURE YOU PASTE YOUR VIDEO FILE URL

this is kind of importanteeeeeee
**extract_nth_frame:** NUMBER OF FRAMES YOU WANT TO BE EXTRACTED FROM YOUR VIDEO

let’s say you have a 4-sec video (30fps)
which in total would be **120 frames**

if you use **extract_nth_frame:** 2
it means, ‘only extracting every second frame’
so from **120 frames,** you would end up with half, **60 fps** in total at the end
it will save you some time, you can stretch out the video, and match the original in post. it will still look cool

if you use **extract_nth_frame:** 1
it means, ‘extracting every frame’
so from **120 frames,** you’d get all **120 frames,** babe!) yep that is right )
but good luck with the wait time though, stay strong! I admire your persistence, or you’re rich AF and have a dope comp. anywayssssss)

Ok, **back to coping your file URL**

Select your file, and then **Copy path**

and THEN we’ve got, prompts!

Here, you put text, like what do you want AI to do with your footage
Every prompt might be different, depending on your shot/scene
I usually experiment and write down keywords, it even works better if you write in full sentences.
Basically experiment and see, if that “look” is what you want/like. Go wild honestly, so many possibilities!

This tutorial really helped me, sharing it in case you wanna check it out! Useful stuff in there.
Will drop some of his prompts down below:
https://youtu.be/Yf2G9ye37ms
These two were best so far from his tut:

[‘cypunk, futuristic, mechanic, dystopia, trending on artstaion, by greg rutkowski, 8k’]
[‘digital painting, nature, kyoto, japan, fantasy, featured on artstation, 8k’]

I AM GOGOKHIA

I AM GOGOKHIA

Comments 0