The Daily Grind: Minimax H3 workflow
Small update to add... Sigh. I discovered the node I was using to display the final calculated prompt text doesn't actually retain that text when you reload an workflow from an older video, which means it's useless when it comes to recovering the prompt to use again or to add to civitai. I'm going to have to replace it with an older text display box that doesn't have the convenient "Copy this" button and release an updated version. The workflow all works as advertised, you can keep using version 1.0, but I'll get you all an update soon in case the prompt thing is something you care about.
What?
When I'm making videos with H3, at least 90% of the time I'm doing one of only a few tasks, each of which needs a little tweaking from the default template workflow. This workflow combines all those tasks into one template that I hope some of you will find easy and helpful to use.
The workflow also includes an option for a Two Stage render pipeline similar in concept to Wan 2.2. Yes, it's more complicated. No, you don't have to do it if you don't want to, this workflow has a perfectly good single stage render process. But if you're a heavy LoRA user, you should give it a try. If your motion/action LoRAs look blurry and messy even at high resolution, or if your nice style and detail LoRA wrecks the audio track in the clip, this option is for you. It lets you use the LoRAs you want in the specific parts of the render process where they will do the most good, without letting them mess up things they're not good at. There are guides and suggestions inside the workflow giving scenarios and recipes for different situations. I won't say you always need the two part render, but I promise there will be times when it's the best way to achieve the result you want.
Why?
I get frustrated by a few things in the standard H3 templates: I don't like having to dive into a subgraph to change a setting for a clip; I think subgraphs make it harder to learn how things work; I hate having to swap templates and models and everything else to switch from text-to-video to reference-to-video; and I want it to be easier to get to the data needed to recreate or remix the video later, like the seed, prompt, and other settings.
How?
I aim to put as many of the things I spend the most time interacting with all in one screen area to limit how much scrolling I need to do. Your work habits will probably vary, so just move things around as you need. If you are new to Minimax H3, you have some downloading to do. I will provide links to all my recommended resources that will get you started, then later you can re-evaluate those choices and swap out some files to suit your own video tastes. If some of the nodes are lit up in red, then either that node is in a group that allows it to be disabled through the toggle switches in the main workflow area, or else it's an important node in the process that can't be skipped and you'll need to download and install it.
After concentrating the most-used stuff around one view area, I don't try to hide any of the moving parts. If you zoom out or scroll around, you will see the whole engine laid out from beginning to end so you can move things around, turn parts on and off, and just tinker however you like. This is not a modern engine block for dummies with a bright yellow dipstick in the middle and a solid plastic cover over all the good stuff. This is a 1960's Caddy with a hood the size of your bedroom and plenty of room to see the whole engine block and track every pipe from beginning to end.
If you give this workflow a try, I hope you find it useful, and post lots of cool stuff here for the rest of us to see.

