This is the most exciting paper I've read in a while.
Alternate title could have been: "One weird trick to increase your depth map inference 200x."
arxiv:
github:
Let's go through the details 🧵 1/9 arxiv.org/abs/2409.11355 github.com/VisualComputin…
Some background. Marigold is a fine-tuned 8-channel version of SD2.1 for depth maps. It concatenates a clean guide image with noisy latent along the channel direction.
It works great but is slow. The authors looked at what happened if they used only a single sampling step. 2/9
Nov 30, 2023 • 11 tweets • 4 min read
“Animate Anyone” was released last night for making pose guide videos. Lets dive in.
