Description
Took some artistic liberties with a cute pic at >>56096
so its possible to use the last frame as the start of a new video and just keep extending it, I think I did that 4 or 5 times to fit the mouth sync. I am sure there are better ways of doings this like infinitetalk and the same technique but I am just trying the hard way before using all available tools at my disposal. Feel free to dub it better and maybe match the emotion better. But its kinda fitting for her to say all that with a smile. She is a little crazy…
Source
not provided yet
We really need like a invoke ai or automatic 1111, but for videos. Using comfyui feels like digging through a dumpster with a rusty spoon.
I’m a supporter of the most generic sounding channel on patreon for AI Knowledge
(central)which gave me a 1 step installer for sageattention for comfyui. It worked after I made an edit to it for my setup. But sageattention is the game changer you are looking for. And generating takes less then 5 mins versus not working or taking 30+ minutes for one GACHA ROLL. There’s an included workflow that amends the last frame of the video as the first to a new one. Allowing you to extend the video in 5 second increments. I am literally on a laptop 4070 with just 8GB VRAM. I don’t want to advertise so I won’t say the name explicitly. PM me if you didn’t read between the lines and can’t find it. I’ll be happy to get another sub for this amazing dude. On a related note I spent the first year of this computer trying different ai stuff and sage attention was something that I couldn’t install on my own after a year of frustration, breaking comfyui, and playing with variables, etc. I wouldv’e paid much more then 5 dollars for the help. And the guy is nice in the comments and helps people that are brave enough to ask. But really I just got lucky and the guy delivered what was promised for once.Pretty good! I’m curious, how are you genning Wan v2.2 on 8gb VRAM?
IIRC that was the “angry” setting on 15. I was just committed to the dialogue I genned on ChatGPT. Even if it didn’t match visually with the emotions. I thought it had some good potential. If I had to do it all over again I would first gen the dialogue, match the voice to the dialogue, use the audio to gen the video with infinite talk lip sync. And see how that looks. And thanks for the compliment. It really gives me some inspiration and ideas about making long form media format. I want to push this tech to its limits. It took like 30 to 50 minutes to gen on 8GB of vram. Didn’t really look at the time but altogether I guess about an hour of editing, compositing, encoding and uploading. Originally it was 39 mb but cloudfail didn’t like that.