Load HunyuanLora From CivitAI

Before wiring this up, know what it depends on: the output type is HYVIDLORA, defined by Kijai's ComfyUI-HunyuanVideoWrapper, and nothing in stock ComfyUI or this pack alone knows what to do with it. Install that wrapper pack first - this node is a drop-in swap for its LoRA-select node, just pulling the file from a CivitAI ID instead of a local dropdown.

What it's for

CivitAI is far and away the biggest public catalog of fine-tunes and LoRAs, Hunyuan's included, and downloading-then-placing a file manually is a real point of friction the moment your ComfyUI isn't running on a machine with persistent, pre-populated storage - a fresh cloud box, an API worker, anything spun up on demand. This node folds the fetch into the run itself: give it an ID, it resolves and downloads the LoRA, and hands it onward in the shape HunyuanVideoWrapper expects.

The inputs and outputs that matter

• civitai_model_id (STRING, required) - "The ID of the model to download from CivitAI." One trap to know about up front: CivitAI's download API works off the model-version ID, not the plain model-page ID visible in the browser URL when you're just browsing a model's landing page. If a numeric ID copied straight from the address bar doesn't resolve, pick the specific version you want and grab its ID from there instead.
• strength (FLOAT, default 1, range -10 to 10, step 0.0001, required) - "LORA strength, set to 0.0 to unmerge the LORA," per the tooltip.
• prev_lora (HYVIDLORA, optional) - chains multiple LoRAs: feed one node's lora output into the next node's prev_lora to stack several before they hit HunyuanVideoWrapper's model loader.
• blocks (SELECTEDBLOCKS, optional) - restricts the LoRA to specific transformer blocks instead of applying it across the whole model. An advanced knob most people leave untouched; it comes from a block-selection node elsewhere in HunyuanVideoWrapper.
• Output: lora (HYVIDLORA) - into the next node's prev_lora, or straight into the wrapper's model loader if it's the only LoRA in the chain.

How to install it

• ComfyUI Manager - search "comfyui-model-dynamic-loader," install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/jax-explorer/comfyui-model-dynamic-loader, then restart.

You also need ComfyUI-HunyuanVideoWrapper installed separately, or this node's output has nowhere to plug in. Neither pack bundles Hunyuan's own checkpoint weights - those are a much bigger separate download, handled through your normal Hunyuan setup.

Common issues & troubleshooting

Nothing to connect the lora output to. That means HunyuanVideoWrapper isn't installed - HYVIDLORA is that pack's own type, not part of stock ComfyUI.

CivitAI ID fails to resolve. Check you're using a model-version ID and not the model-page ID first; if that's not it, the model may simply have been removed. CivitAI's catalog loses a real share of older uploads to takedowns and license-driven removals over time, so a dead ID from an old guide isn't unusual.

LoRA loads but has no visible effect. Confirm on the CivitAI page that the LoRA is actually built for Hunyuan and, if relevant, for the specific Hunyuan version you're running - a mismatch doesn't error, it just does nothing. Also check for a trigger word requirement in the prompt.

Worth knowing before you invest time here: Hunyuan's community LoRA ecosystem has thinned considerably since Wan 2.2 became the default for open local video work - fewer new LoRAs get published for it, and fewer guides get written. If you're starting a new project rather than maintaining an existing Hunyuan pipeline, it's worth checking whether what you need already exists for Wan, where the library is deeper and still growing.