I created a video from animations generated by two AI animation generators -- both PlaiDay and PIKA LABS:
SAME VIDEO FACE
The new PlaiDay software allows one to generate an AI video in Discord, and then swap out the face in the video. So I swapped out the face with one that I had generated in ArtBreeder in 2021:
I swapped the the original face in the PlaiDay videos
After I first generated AI videos in PlaiDay, I swapped out the face. Below are some examples of the PlaiDay videos, all after I regenerated them with the same face:
PlaiDay
prompt: "a clown face crying"
prompt: "a death metal guitarist eating flies"
prompt: " a country and western singer on state"
prompt: "newscaster on TV"
Refried in
PIKA LABS
I then used EzGif to make animated GIFs of the PlaiDay animations (like the ones above), and uploaded those GIFs into PIKA LABS to generate more AI videos (PIKA LABS just accepted the first frame of the animated GIF). Note that there is less motion in the PIKA LABS videos, such as in the examples below:
Switching out the face made a huge difference. The video animations had more substance for being about a particular person -- and then being about the same person made the compilation video even more coherent.
- Matt Wolfe made a YouTube video about PlaiDay: Create AI videos Using YOUR Face!
Matt's video also alerted me to Moonvalley.AI, another free AI animation program on Discord:
prompt: "Party at the hackerspace"
Video2Video AI
WHEREAS PlaiDay and MoonValley.AI allow one to generate a video from only TEXT,
AND PIKA LABS allows one to generate a video from text and/or an IMAGE,
ModelScope Video2Video allows one to use AI to alter a whole VIDEO!
I uploaded all the PlaiDay videos and altered them with ModelScope Video2Video, using the prompt: "robot," -- then I remade the video at the beginning of this blog post:
The ModelScope results were hit or miss...
The best result from a PlaiDay video
altered with ModelScope, prompt: "robot"
... however just the idea that I could alter a WHOLE video with AI is impressive.
It takes about a half hour to alter 3 seconds of video with AI, which I did remotely using the beefy GPU computer at Quelab.
Imagine how long it would take to alter a whole movie with AI. But Hollywood can probably do that right now
No comments:
Post a Comment
Note: only a member of this blog may post a comment.