Hunyuan Video LoRA Loader - Load From Cache

HunyuanLoadFromBlockCache is the quiet sibling in the PixelFun pack, and it only makes sense next to the other two. Where Edit Blocks is the editor, this is the replay button: it loads a Hunyuan Video LoRA using whatever block settings you already saved for it, with none of the 20 toggles cluttering your graph.

The workflow it's built for is a two-phase one. Phase one: load the LoRA in HunyuanLoadAndEditLoraBlocks, toggle double blocks until the output is right, and let save_settings write your choices to disk. Phase two: swap in this node, which reads those settings back and applies them. Your graph stays clean, and every future run of that LoRA gets the same surgical treatment. It's the difference between keeping a recipe in your head and keeping it in a file.

How it works

When you save settings in Edit Blocks, they land in a cache directory inside the pack:

ComfyUI/custom_nodes/ComfyUI_PixelFun/cache/

Each LoRA gets its own file named after it - kerbal30_blocks.yaml, for instance. This node loads that YAML, merges it over the defaults (all 20 double blocks enabled), then filters the LoRA's tensors down to whatever blocks are switched on before handing them to ComfyUI's standard LoRA apply machinery. Same filtering trick as the editor, just no UI.

The inputs and output

This one is almost boring, which is the point:

• model - the base model to patch (MODEL in).
• lora_name - the LoRA from models/loras.
• strength - global strength, default 1.0, -10 to 10. In practice this is the only field you'll touch day-to-day.
• use_single_blocks - an all-or-nothing switch for the 40 single blocks (default off). Overrides whatever the cache says.
• Output: the patched model.

The gotchas, such as they are

• No cache file means nothing special happens. If you've never saved settings for a LoRA, this node just loads it with everything enabled - functionally identical to a stock LoraLoader. If it "doesn't do anything," check that you actually ran Edit Blocks with save_settings on first.
• Settings are keyed by filename. Swap in a new .safetensors with the same name and you inherit the old block config. Rename to be safe.
• Single blocks aren't cached. Only the double-block toggles are saved, so use_single_blocks here is a per-load decision, not part of the recipe.

Installing it

Same as every node in the pack - it's one git clone away:

cd ComfyUI/custom_nodes
git clone https://github.com/PixelFunAI/ComfyUI_PixelFun

Restart ComfyUI, or search "Hunyuan LoRA Loader Nodes" in ComfyUI Manager. There are no extra dependencies and no model downloads; the pack's only real requirement beyond ComfyUI itself is yaml, which is already bundled.

Honest take: this node is only worth installing if you're actually doing block surgery on Hunyuan LoRAs. If your Hunyuan workflow is "load LoRA, set strength 0.8, render," you don't need it - and you'd never notice it was there. But the moment you've dialed in a block config you like, the cache it reads from is what makes that config reproducible instead of a screenshot and a prayer.