Long time lurker, first time poster. Have been playing around with some AI recently and thought I'd share some of my results. The main frustration is the 5 second(ish) bottleneck of wan2.2, have stretched it out to 6 seconds but, like others, am waiting for the tech to progress in order to produce longer vids with satisfying after shots.
Not really an expert, not fine tuning loras like some, just hacking my way through comfyUI. These 4 gifs come from similar prompts, will do more with more variety if response is remotely positive.
Oh wow, replies! Thanks so much. Here's 4 more duo gifs.
I don't save my prompts but I can tell you that in flux I specify a "tall glass box with a wide pipe at the top" for the tank. Then in wan2.2 I say "very thick gunge pours out of the pipe". I do also say "sticks to them" to prevent the gunge from magically disappearing. The gunge tank with the tank at the top was really difficult to generate and took a lot of trial and error so I won't be bothering with that again
For the lighting I've noticed that the scene will take on a tint based on the colour of the clothing which is something I haven't been able to work around. I could probably do more prompt engineering to get around this but I kinda want to keep the scene minimal so I just say "bright room".
Cos it's a 2 step process (image gen > image to video) I generate 10 pics at once and basically select the ones where the pipe is adequately wide and positioned correctly. Then I run it through image to video and let the AI physics do its thing. This saves on generating a ton of videos which is the slow bit.
Anyhow J-girls in gunge tanks is basically the limit of my creativity so I'm not gonna keep posting on here. I'll be back when open source progresses and longer videos are possible.
Love you, too
