Load Lora From ComfyOnline

The pack this node ships in - comfyui-model-dynamic-loader - describes itself in its own README as built "for comfyonline dynamic loader," and this node is the clearest example of what that means. It isn't a general-purpose LoRA fetcher; it's plumbing for ComfyOnline (comfyonline.app), a paid ComfyUI-cloud service, and the ID it expects is one from that platform's own model library - not CivitAI, not HuggingFace.

What it actually is

Set aside the ComfyOnline wiring for a second and it's structurally the same as this pack's other Hunyuan and Wan LoRA nodes: it produces a HYVIDLORA object, the custom type Kijai's ComfyUI-HunyuanVideoWrapper defines and consumes. That means two things have to both be true for this node to do anything - you need HunyuanVideoWrapper installed for the output to plug into, and you need a comfyonline_model_id that actually resolves against ComfyOnline's own backend, which almost certainly means having an account and models uploaded or licensed there. Outside that platform, this node has nowhere to get its data from; the README documents no fallback and no generic endpoint.

Worth knowing going in: ComfyOnline itself gets a mixed reception in the community. Some people use it for cheap cloud training and generation and share workflow guides off the back of it; others flag that ownership is opaque (WHOIS records redacted, registered out of Beijing per user reports) and complain about it showing up as unsolicited ads across ComfyUI communities. None of that is a knock on this specific node - it just means "does the ID field even work for you" depends entirely on whether you're already inside that ecosystem.

The inputs and outputs that matter

• comfyonline_model_id (STRING, required, default empty) - "The ID of the model to download from ComfyOnline," per the node's tooltip. Where this ID comes from isn't documented in the pack's README; presumably it's whatever identifier ComfyOnline's own library assigns a model you've uploaded or selected there.
• strength (FLOAT, default 1, range -10 to 10, step 0.0001, required) - "LORA strength, set to 0.0 to unmerge the LORA."
• prev_lora (HYVIDLORA, optional) - chains multiple LoRAs together, the standard HunyuanVideoWrapper stacking pattern: feed one node's lora output into the next node's prev_lora.
• blocks (SELECTEDBLOCKS, optional) - restricts the LoRA to specific transformer blocks rather than applying it across the whole model, an advanced knob most people leave alone.
• Output: lora (HYVIDLORA) - into prev_lora on the next node, or directly into HunyuanVideoWrapper's model loader.

One cosmetic note: the node's display name in the UI carries a stray leading space (" Load Lora From ComfyOnline") - harmless, just don't be thrown by it when you're searching the node list.

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'll also need ComfyUI-HunyuanVideoWrapper installed for the HYVIDLORA output to connect to anything, and a working ComfyOnline account for the ID field to resolve to a real file.

Common issues & troubleshooting

The ID field just doesn't fetch anything. If you're not signed up with ComfyOnline, that's expected - this node has no visible fallback to a public source, unlike its CivitAI- and HuggingFace-sourced siblings elsewhere in this pack. If you want a Hunyuan LoRA and don't use ComfyOnline, reach for LoadHunyuanLoraFromCivitAI or LoadHunyuanLoraFromHF instead.

Nowhere to plug the output. Install HunyuanVideoWrapper - HYVIDLORA is its type, not native ComfyUI's.

You found a Hunyuan LoRA at all. Worth setting expectations: Hunyuan's community LoRA ecosystem has thinned out considerably as Wan 2.2 pulled most of the attention and tooling investment. If you're starting a new video project from scratch in 2026 rather than maintaining an existing Hunyuan pipeline, it's worth checking whether Wan covers what you need first - the LoRA library there is deeper and still growing.