Load Model Meta (QQ)

Most of the time, a checkpoint loader is a checkpoint loader: pick a file, get a MODEL. LoadModelMeta is the version that quietly attaches a string of metadata to the model it loads - specifically the full path of the checkpoint - because the node it feeds, LTX2LoRASelect, needs to know exactly which base model a LoRA chain was built on. It's not a node you reach for on its own. It's the first link in the pack's LTX LoRA-merge pipeline, and it exists to solve a bookkeeping problem no one else bothered to solve.

It ships in siraxe/ComfyUI-WanVideoWrapper_QQ under the KJNodes/ltxv menu, sitting next to the LTX2 LoRA tooling. If you've ever wanted a loader that remembers provenance, this is the blueprint.

How it works

Under the hood it's a thin wrapper over ComfyUI's standard load_checkpoint_guess_config - same model loading, same behavior, no new formats. The difference is one line: after loading, it stashes the checkpoint's full path into model.model_options["checkpoint_name"]. When the model travels forward through a chain of LTX2LoRASelect nodes, that metadata rides along inside the model object, so the merge stage (LoraMergeLTX) can reconstruct exactly what the chain was built on. The metadata survives as long as the model tensor does - which is the whole point; ordinary loaders drop that information on the floor.

One quirk worth noting: the checkpoint name is stored as the full filesystem path, not the friendly dropdown name. That's deliberate - the LoRA merge needs an unambiguous file reference. If you move your model files after loading, a saved workflow may point at a stale path, which is worth remembering when you share workflows built on this pipeline.

Inputs and outputs that matter

• checkpoint_name - the dropdown of checkpoints in your models/checkpoints folder.

Output: a single MODEL, wired to the first LTX2LoRASelect in your chain. Note it does not output CLIP or VAE - if you need those, load them separately with a normal loader.

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/siraxe/ComfyUI-WanVideoWrapper_QQ.git

Restart ComfyUI (or ComfyUI Manager → "WanVideoWrapper_QQ"). No extra dependencies - this node uses only ComfyUI's own loading code, and the pack's requirements.txt is empty. If you're upgrading from before the pack's v1.3.4 rename to ComfyUI-SA-Nodes-QQ, delete any old wanwrapper_qq folder to avoid import collisions.

Common issues

Realistic failure modes are few. If your checkpoints dropdown is empty, your models aren't in the standard folder - a layout issue, not a node issue. If you wire a MODEL from a different loader into LTX2LoRASelect, the metadata never gets attached and LoraMergeLTX will later complain about a missing chain; that's the pipeline telling you to start with LoadModelMeta. And be mindful that this is LTX-pipeline plumbing - using it in a plain Wan or SD workflow gains you nothing except an unneeded dependency on the pack.