SaveLatentMXD

The save half of Maxed Out's latent save/load family, and the node that makes the Wan 2.2 two-stage workflow feel sane. You run a high-noise pass, save the latent, and later load it back with a MXD loader - no need to keep the first model loaded while you agonize over which drafts to finish. This is the "save" button for that whole loop.

Why bother? Wan 2.2 splits generation into a high-noise model (motion, composition) and a low-noise refiner (detail). The smart way to work with that is to batch out lots of cheap starts, then spend expensive refinement only on the ones worth it. If you don't save the latents, "finish it later" means re-running the first pass. If you do, you're just picking files. The community has been asking for exactly this since Wan 2.2 shipped - it's a small thing that saves real VRAM and real time.

How it works

This node writes your latent as a .latent safetensors file under ComfyUI/input/latents/, embedding the source workflow as metadata so a MXD loader can reconstruct what you did. The loaders read that metadata back and hand you the tensor plus your sampler settings, which is what turns "resume a run" from a prayer into a dropdown selection.

Note this is the older, stripped-down key: it takes just the tensor and a filename prefix. The newer Save Latent variants also stash positive/negative conditioning in a sidecar - this one keeps it simple and writes the bare latent.

Inputs

• samples - the LATENT tensor to save.
• filename_prefix - the base name; files land in input/latents/ with a counter appended, so repeated saves don't collide.

That's it. It's an output node, so it terminates the graph.

Installing it

Part of the ComfyUI-MaxedOut pack. ComfyUI Manager search Maxed Out, or:

cd ComfyUI\custom_nodes
git clone https://github.com/Maxed-Out-99/ComfyUI-MaxedOut.git

Restart ComfyUI. No extra dependencies, no model downloads.

What to watch for

This is a legacy key - Maxed Out renamed the latent family and keeps the old ones so existing workflows still load. If you're setting up fresh, the current Save Latent nodes from the same pack save conditioning too, which makes the load-back side much more complete. And remember the save location: input/latents/, not your output folder. Loaders pick files from there, so a latent you save with a different pack's saver may not show up in the MXD dropdown.