Checkpoint Loader_v2 (QQ)
The checkpoint loader with a PowerLoraLoader-style UI
- MODEL
- CLIP
- VAE
ComfyUI's built-in CheckpointLoaderSimple is fine. It loads a checkpoint. You'll never be lost using it. What it doesn't do is fit into a tidy "power" workflow where you toggle models on and off from a dropdown without rewiring your whole graph. CheckpointLoader_v2 from the "(QQ)" pack is the answer to that specific itch: a checkpoint loader styled like PowerLoraLoaderV2, where the model choice is a widget you flip rather than a wire you re-route.
It comes from siraxe/ComfyUI-WanVideoWrapper_QQ, sir_axe's Wan toolpack. The UI convention it copies is the rgthree-style flexible loader - the author credits rgthree in the README, and you'll recognize the DNA instantly if you've used the PowerLoraLoader family.
How it works
There's a wrinkle you'll notice immediately: the node's visible inputs are dynamically generated. The schema exposes a flexible input system (the pack uses the FlexibleOptionalInputType mechanism), which is why a "loader with no required inputs" still manages to output a MODEL. In practice you pick a checkpoint from the model dropdown it renders, enable it, and hit run. The node scans its dynamic inputs for the one that's turned on with a valid filename, resolves it via ComfyUI's checkpoints folder (the same folder the built-in loader reads), and loads it with ComfyUI's standard load_checkpoint_guess_config. Output is the usual triple: MODEL, CLIP, VAE.
The important grounding: this is a thin wrapper over ComfyUI's own loader. It is not loading anything exotic, and it doesn't invent new model formats. The value is entirely ergonomic - you can sit it in a graph, swap checkpoints from the widget, and keep everything else wired.
What you actually set
- The checkpoint dropdown (the dynamically-added model input) - enable one and pick your file.
- Nothing else, really. It doesn't even take a clip-skip or VAE override.
Outputs: MODEL, CLIP, VAE, wired exactly like the stock loader's outputs - model to your sampler, CLIP to your conditioning, VAE to your decode.
Installation
cd ComfyUI/custom_nodes
git clone https://github.com/siraxe/ComfyUI-WanVideoWrapper_QQ.git
Restart ComfyUI, or install via ComfyUI Manager (search "WanVideoWrapper_QQ"). No extra dependencies - this node leans entirely on ComfyUI's built-in model-loading code, and the pack's requirements.txt is empty. If you have an old wanwrapper_qq folder from before the v1.3.4 rename to ComfyUI-SA-Nodes-QQ, delete it to avoid import collisions.
Common issues
The classic failure is running it without selecting a model: you get "No checkpoint model selected or enabled," which is the node telling you the dynamic widget wasn't configured - open the node, enable the dropdown, pick a file. If the dropdown list is empty, your checkpoints aren't in ComfyUI's models/checkpoints folder, which is a folder-layout problem, not a node problem. And because it's a wrapper, it inherits whatever quirks your checkpoint has - a model that fails to load in the stock loader will fail here identically. Worth knowing if you're debugging: there's no extra magic layer to blame.
Inputs (0)
No inputs
Outputs (3)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| CLIP | CLIP | — |
| VAE | VAE | — |