HunyuanVideo TextEmbeds Load

The other half of the save/load pair, and by itself it's about as simple as a node gets: one dropdown, one output. Point it at a set of embeddings previously written by HyVideoTextEmbedsSave - the tooltip on the embeds field spells out exactly where it looks: output/hyvid_embeds, ComfyUI's own output directory - and it hands you back a ready-to-use hyvid_embeds object, the same type every sampler node in this pack expects.

Why this exists: Hunyuan's text encoding runs through an LLM, and that's genuinely one of the slower, more VRAM-hungry steps in a generation. If your prompt isn't changing between runs - you're tuning steps, seed, LoRA strength, guidance, whatever - re-encoding the same text every single time is pure waste. Encode once with HyVideoTextEncode, save it with HyVideoTextEmbedsSave, then replace that pair of nodes with this one for the rest of your iteration session. Nothing downstream needs to know the difference; hyvid_embeds is hyvid_embeds regardless of which node produced it.

There's no strength, weight, or blend control here - this node doesn't modify anything, it just loads. If you want to combine two saved prompts or blend between concepts, that's a different tool (HyVideoPromptMixSampler takes two separate hyvid_embeds inputs and blends them at generation time - this node can happily feed either slot of that sampler with a saved embed instead of a freshly-encoded one).

It's also worth reaching for during batch runs. If you're queuing up a big seed sweep or a long unattended overnight batch against a fixed prompt, loading a saved embed means the LLM text encoder never has to sit resident in VRAM alongside your diffusion model and VAE for the whole run - one less thing competing for memory across a hundred queued generations, and one less place for a batch to slow down or stall partway through.

Installing the pack: search "HunyuanVideoWrapper" in ComfyUI Manager, or clone it by hand - cd ComfyUI/custom_nodes && git clone https://github.com/kijai/ComfyUI-HunyuanVideoWrapper, then pip install -r requirements.txt inside the folder, then restart. That pip install step catches people out more than anything else in this pack: Manager-only installs without it tend to leave nodes missing from the search box.

The one thing that'll trip you up: this node's dropdown only shows files that already exist under output/hyvid_embeds/. If the list is empty, it's not broken - it means you haven't run HyVideoTextEmbedsSave yet, or you saved to a different filename_prefix/subfolder than this node is looking in. Run a save first, then come back and refresh the node to see it in the list.