TL;DR: A LoRA for perfect feet, but it's tricky: check that left and right foot aren't both "right feet" and count the toes – a clean prompt fixes it. Bonus: thanks to the new Cine Timeline, all three shots in one run – same person, same moles.

----

So, dear Tarantino foot fetishists and those about to become one,

good news for you! Today I'm sharing a LoRA for perfect feet. Like all the other LoRAs for MiniMax H3, it's just a little helper that gets the toes into the right places – clean prompting does the rest. Because: during testing I got two right feet in the shot twice, both "big toes" on the right side, so not the anatomically correct mirror-image setup … and once even six toes. Thought we'd left that behind ages ago, but yeah, yeah, it might be this LoRA's fault. I managed to clean all of it up with a proper prompt. Keep an eye out for that stuff. In general, this LoRA can also help push the camera to a full shot / full body, so you get to see the whole figure …

The example videos, though, use a new Cine Timeline custom node I'm using here for the first time. It doesn't just do I2V with input images as keyframes, it also fully supports T2V – pure prompting, but with an exact second-by-second breakdown – plus various editing tools like "Cut", layer management and so on, Adobe Premiere style … And here's the real point: the three shots in each video come out of ONE single run, not three separate generations glued together afterwards. That gives you a rock-solid consistency you usually just don't get otherwise: the moles stay exactly where they are, the face, the body, the skin, every detail stays put – it really is the same person in different poses and camera angles, not three lookalikes. I've just added the node to the GitHub repo along with a bunch of other updates, among them a search function in the CLIP Text Encode node. Still BETA, not a final node version. So please don't yell at me right away if something doesn't run quite the way it's meant to. I'll test everything properly over the next few days … for now, this is a work in progress. :) Download the new version straight into your ComfyUI custom_nodes folder – my repo is officially registered with ComfyUI, but the versions there lag behind:

https://github.com/PolyhedronAI/ComfyUI-PolyhedronLoRAStack

(There's also an example workflow with the "fox" for MiniMax H3.)

All the workflows for the girls shown here, plus every LoRA I used, are available for download on my Patreon page as usual. If you like what I do, it would be great if you could actively support me …

https://www.patreon.com/c/polyhedron_ai

Gear: 5060 Ti 16 GB VRAM // 32 GB RAM

Other than that: go test it! Dare more feet. :D

Foot Fetish Frank