Get IC-LoRA Parameters

GetICLoRAParameters is a tiny node with one genuinely useful job: it reads the safetensors metadata off an IC-LoRA and tells you - or rather, tells LTXVAddGuide - how that particular LoRA wants its guide images handled. Some LTXV IC-LoRAs were trained with a "reference downscale factor" baked into their metadata, and if you feed such a LoRA a guide without accounting for that factor, the conditioning lands at the wrong latent scale and your pose/depth/reference guidance comes out misaligned or mushy. This node extracts that factor so the pipeline can compensate.

The "IC" in IC-LoRA stands for in-context - Lightricks' family of control LoRAs for LTX Video that condition on reference images (pose, depth, canny, or plain subject references) by stacking them as in-context frames. They're one of the things that made LTX's speed tier actually usable for VFX-style work. And because each LoRA can carry its own trained-in downscale factor, the built-in nodes added this little metadata handshake rather than hardcoding one value for all of them.

How it works

You connect it to the output of a LoRA Loader - specifically the model output, not the model that loaded it elsewhere. The node pulls the lora_metadata attachment off that model, looks for a key ending in reference_downscale_factor, and rounds it to an integer (defaulting to 1 if there's no such key or it's malformed). It packages that as an IC_LORA_PARAMETERS object. Straightforward, but note the important detail: it must be the direct LoRA-loader output, because that's where ComfyUI stashes the safetensors metadata during the load.

Inputs and output

• iclora_model - direct output from a LoRA Loader for the specific IC-LoRA (per the tooltip: it extracts metadata from that LoRA).
• Output: iclora_parameters - connect to LTXVAddGuide's iclora_parameters input.

Where it fits

Ships with ComfyUI core, part of the LTXV node family. No model files of its own. The practical rule from the tooltip: connect it to LTXVAddGuide if the LoRA requires special handling - in practice, most IC-LoRAs ship with a factor of 1 and this node changes nothing. It exists for the ones that don't, and when you hit one (features misaligned, guide looking downscaled), this is the official fix.

Common issues

The failure people hit is wiring it wrong: feeding it a model that has already passed through other nodes, or a different LoRA than the one LTXVAddGuide is using - the parameters then describe the wrong LoRA. Also, per the docs, when you chain multiple LTXVAddGuide nodes, each one uses only the parameters connected to it, so if you have several guides, give each its own GetICLoRAParameters wired to the corresponding LoRA. And don't expect the node to tell you a factor exists - if the output is 1, the LoRA just doesn't need special handling, and that's a valid, boring answer.