My first stumbling attempts at AI videos

[Spindrift]

After a lot of trial and error I've finally managed to get my 3 year old GPU to locally generate some semi-decent AI video. But I'm still learning the process so the quality is not too good at this point. Messed up aspect ratios resulting in scrunched or stretched videos, jerky movements due to bad setup, and low resolution resulting in murky details. I am posting in gif format because I am too cheap to pay for a video hosting service, but this site won't allow gif file sizes greater than 32MB, so most video resolutions will be 640x640, maybe 768x480 if the video is short. Constructive critiques, tips, hints, tricks are always welcome.

[Spindrift]

Maybe you recognize her?

[dubiouskey]

Not bad! Good first attempts.

[Spindrift]

Not bad! Good first attempts.

Thanks! Still trying to work out aspect ratios and longer play times.
Here's a few more.

[Hand-Holder]

If this is stumbling, than next month you are the King, great stuff

[Spindrift]

If this is stumbling, than next month you are the King, great stuff

Thank you! Still working on "in-hands" for now. Hoping to work up to "hand helds" soon. Also trying to get a little more life-like motion in the videos.

[Queen Skadi]

So what are you using to generate these and how easy is it to set up on your computer? Have tried other video generators but even relatively tame prompts seem to get rejected

[Spindrift]

So what are you using to generate these and how easy is it to set up on your computer? Have tried other video generators but even relatively tame prompts seem to get rejected

I will never generate images or video on any online AI site after the puritanical hellscape its become. I generate all my stuff on my own computer, even though its old and takes a while to generate images and especially videos. I use Stable Diffusion tools such as WebUI Forge and SwarmUI mostly. I used ComfyUI for a while, but it kept breaking whenever I updated it or tried to use a workflow made by someone else. Right now I am using the Image2Video model of Wan2.2 to convert my previously generated images into video. A slow process for my RTX 3080 GPU!

Some more Wonder Woman vidoes. You can see I am trying to do multiple things at once, SW image to video along with a recognizeable celebrity face. The preceding Wonder Woman you might recognize as the star of the "Tomb Raider" movies. This Wonder Woman is the star of the most recent DC movies. I'm still trying to get the giant hands to touch where I intend them to.

[Spindrift]

An elderly retiree being interviewed about how getting a SW changed the quality of his life.

[Spindrift]

Incorrect aspect ratio is causing this slender little lady to appear much thiccer than she actually is. At least the movements are improving.

[Spindrift]

Same prompt, different ccelebrity.

[Spindrift]

Meanwhile, in Japan...





The AI can't seem to understand that the jade carving isn't made of rubber!

[Spindrift]

And while we're on the subject of Japan, a few image to videos from my Bonsai series. Inspired by the first "Evil Dead" movie.

[Spindrift]

Slow improvement toward handheld. Notice in the 3rd video the AI thinks the threaded bolt is a bendy straw!

[Spindrift]

Another "workshop" video where the AI seems to think a threaded steel bolt is some kind of spring.




And one where we see some unexpected ragdoll physics!

[Spindrift]

Some controversial religion themed videos.

[Hand-Holder]

Nice priest, just helping them synchronize... Nice fellow, at least for now !!!