This is my personal-use workflow I put together with what I've learned while using Minimax H3. What it centers around is using a latent upscale pass into a second sampler, which has two advantages:<ul><li>It allows you to quickly test out lower-resolution video segments until you get one you like.</li><li>The second sampler helps clean up and further detail the video, including the motion.</li></ul>What's notable about the workflow is that I've pieced together a way for context and video continuation to work while using that two-stage sampler.This took me a while, as using two pieces of context between samplers wasn't something the nodes normally intend for. As a result, a few lines of code modification on one of the nodes is needed (which is detailed as a note in the workflow, allowing "pins" on the latent to be overwritten).

Files with SHA256: 927f7aba3d599528d60efa91fcf592cc848dcb17787826295125583029304120