I've been working on my video editing game, and seeing how these short Wan2.1 clips can be put together. And creating more specific details in the scene too. Here's my first bigger try at that, a gunge tank scenario.

"Bigger" still means <30 seconds, no sound, but I am hopeful about pushing this further. I'm sure better models in the future will allow more too ... There are two clips bundled together here, one longer than the other. Let me know what you think!

This was done by special request from MessMan83 -- thank you sir!!

Good stuff and you beat me to it - I'd just made a decent 5 second gungetank clip the other day using T2V.

How did you get the woman to look consistent - was this by re-using end-frames as start frames?

Thanks!! And yep that was my basic strategy. Sometimes using a start frame, sometimes using a start and an end frame. (Which is possible now with some custom nodes by a ComfyUI guy called Kijai.) It's still all a bit by chance, pick the best looking results and run with that. Honestly it might be faster just to spin up a small lora for a character, I might try that to see if it works.

My process went basically like this:
1. Use Wan to generate some gunge tank clips with text-to-video, decide that wasn't working
2. "Draw" an empty gunge tank with a stool in it, in Gimp, and process that into something realistic-looking in Flux
3. Use that as a starter image, use Wan image-to-video to generate some short clips of a woman sitting down on the stool. Try a few times till one didn't violate every law of physics
4. When I had a clip of a woman sitting down, take the last frame, back to Gimp and Flux to draw the closed plexiglass half-door in front of her
5. Then, Wan image-to-video with a start frame (woman sitting, no door) and end frame (woman sitting, closed door), to get a snippet where she's realistically closing it. Note how there's a cut there, this was the hardest bit to generate.
6. Wan image-to-video, sitting-to-gunged, for the money shot.
7. Then use Gimp and Flux again to "draw" the right kind of splatter on the plexiglass panels, and use those results as an end frame for another I2V pass.
8. Wan image-to-video, fill in a little bit more "clean posing" in the middle

Stitch all that together in OpenShot with some basic crossfading, and there you go.

The shorter, simpler clip was a smaller simpler chunk of that workflow.

also, here are some gif "bonus tracks," shorter clips that I didn't work on so much. Mostly where I decided I wasn't getting enough of the tank scenery that I wanted. But still fun! And crisper since I didn't edit and overwork them so much.

These are great! Keep up the slime work. The girls are very cute and the reactions are sublime.
I haven't created in a while, but did compile some of my better Kling clips into a video. It's on my profile page if you'd like to check it out.

Incredible, game-changing work!!! Love it!!!

Great progress being made. I'm a big fan of WAN2.1 but leaving my testing and training until my new GPU arrives. I've had loads of issues with the 5090 running locally. CUDA, torch etc are not very stable right now. Instead, I've decided if I'm going to go through the pain, I might as well go large. I've got the new A6000 pro blackwell 96Gb on order which should arrive mid may.