Documentation

Understand

LoRAs

Add a style, a character or a motion to a model with a link to a .safetensors file.

A LoRA is a small weights file, trained for one model. It teaches that model a style, a character, a product or a camera move, without switching models. It almost always comes as a .safetensors file.

In iNodes, about a hundred nodes take a LoRA. Their name says LoRA or with LoRAs: FLUX.1 [dev] with LoRAs, Flux 2 LoRA, Qwen Image Edit Lora, Wan 2.2 14B I2V LoRA and the LTX-2.3 Lora nodes, among others.

The two settings

  • LoRA URL: the link to the LoRA file.
  • LoRA Scale: how strong the LoRA is, 1 by default. Lower it if the effect takes over your image, raise it if you can barely see it. The range depends on the model: 0 to 2 on the LTX-2.3 Quality nodes, for example. Model info tells you.

A node takes one LoRA. On the LTX-2 and LTX-2.3 Lora nodes and on the MiniMax H3 LoRA nodes, the link is required: without a LoRA, use the same node without "Lora" in its name. The cost is the same with or without a LoRA.

You don't upload the file: the model provider downloads it from your link, without your account. So the link must meet three conditions.

  1. A direct link to the file. It starts with https:// and points to the .safetensors itself. A path on your disk doesn't work.
  2. A public file. It must download without signing in. A "gated" Hugging Face repository, whose terms you must accept, fails even after you accepted them: your account stays on your machine. The same goes for a Civitai model reserved for signed-in members.
  3. A LoRA trained for this model. An LTX-2.3 LoRA goes on an LTX-2.3 node, a FLUX.1 [dev] LoRA on FLUX.1 [dev] with LoRAs. On another model, it is refused or has no effect. Some models also cap the file size: 3 GB on LTX-2.3.

On Hugging Face, open the repository's Files and versions tab and click the .safetensors file. You can copy the address of that page, the one with /blob/ in it: the iNodes app turns it into the download link, /resolve/. The repository link alone, without the file, isn't enough.

When the node refuses

The app checks the link before sending anything and shows the problem on the node:

  • This model needs a LoRA: paste its link in LoRA URL: this node requires a LoRA and the field is empty.
  • LoRA URL must be an http(s) link to the weights file: the field holds something other than a web link, often a path on your disk.

If the link looks right but the generation fails, first check that the file downloads in a private browsing window, signed out. That is the most common cause.

Updated 8 Nov 2026

LoRAs | iNodes