Dear Beach Boys and Hula-hoop Girls,over 34 °C here in Germany today — and I'm already a fully hard-boiled egg. But okay! The Tequila Sunrise is parked next to the keyboard, the cat has melted into a black-and-white puddle on the laminate… so it's time to upload a new LoRA: this time, The Ocean.Trained on stills, so T2I; I already used them for a Flux.1 LoRA, but I figure it still has an effect on the WAN 2.2 T2V model too — maybe just nudging the overall lighting mood a bit, or rendering a rocky, sandy beach in more detail… not a real game-changer, but definitely nice to have.For now it's only the low-noise here; the high-noise goes up on my Patreon first, and later here on <a target="_blank" rel="ugc" href="http://civarchive.com">civitai.red</a> too. // You know the drill. Take it as an incentive to actively support my work… (You can still drop the low-noise model into the high-noise path too, and then you'll get what you want as well) <a target="_blank" rel="ugc" href="https://www.patreon.com/c/polyhedron_ai">https://www.patreon.com/c/polyhedron_ai</a> Anyway! This is my first shot at showing you not just stills but some first short clips… I put it off for a long time, but by now I think I've grasped the basic principles of WAN 2.2 well enough and can follow how the generation works…Right now I use the classic KSampler for my work — well, two of them for the HIGH/LOW split — plus the MoE sampler, which does all of that automatically in one go, via a boundary at 0.875 for T2V.Classic trouble spots in video generation that don't really come through much here: as usual, nipples potentially showing through the wet fabric — see my last WAN 2.2 LoRA "Wet Shirt" — then the whole consistency thing: pupils, eyes that track the camera cleanly; then all the imperfections like moles that move along properly with the model's motion; then the hands, the fingers, the sweat on the skin too; and then the background animations like the waves breaking, and so on and so forth.Looks all clean enough to me. :)Sure, not perfect; it can always be better. Everything can always be better… But this is a 65-frame setup (at 768 × 768 px) with a simple ESRGAN upscale up to 1024 × 1024 (there'd probably have been even more in it^^). On my 5060 Ti with 16 GB, on 14B fp8 models; only 8 steps with the Lightning T2V accelerators… and it still doesn't look like plastic anymore… Takes around 12 minutes — those are easy to invest for a clean video…So that's the current state of the art. All the example videos, with every LoRA used, with the workflows included into the video files, are — as always — on my Patreon. If you can still move your index finger in this heat: click me!Have a good time. Have a beer for me. &lt;3 Frank

Files with SHA256: 093a08e6f7cb89a974ddb2eebc946930bdcf50d8f4582efb3452a29e22ef6660