Checkpoint Info Loader

Most checkpoints you download from CivitAI or similar sites come with example preview images, and those PNGs usually have more baked into them than pixels - ComfyUI can embed a full workflow graph directly into a PNG's metadata, so dragging one of those example images onto your canvas loads the exact workflow (samplers, settings, prompt) that produced it. Normally you get at that by manually downloading the preview and dragging it in yourself. This node does the same lookup from inside a workflow: point it at a checkpoint, and it pulls the associated preview image plus whatever workflow and metadata are embedded in it, as outputs you can wire up directly.

What it does

checkpoint picks the source file from a dropdown. workflow_index (default 0, range 0–100) matters when a checkpoint has more than one associated example - pick which one you want by its position in the list. Outputs: preview (an IMAGE, ready to plug into a Preview Image node or anywhere else an image output goes), workflow_json (the embedded ComfyUI workflow graph as a string), and metadata_json (whatever other generation metadata - sampler, steps, CFG, the works - travels with that example, as a string).

Why this is genuinely useful

The obvious case is exactly what it sounds like: you've got a checkpoint on disk and want to know how its author actually intended it to be used, without leaving ComfyUI to go dig through a browser tab. workflow_json gives you a starting point workflow for a model you might otherwise be guessing settings for. metadata_json is the faster read when you just want the numbers (sampler name, steps, CFG) without reconstructing a full graph.

It's also the natural companion to EmbeddingInfoMetaDownloader elsewhere in this pack, which is built to bulk-fetch exactly this kind of preview and metadata for embeddings from a remote repository - this node is the read side for checkpoints already sitting locally with their previews attached, rather than the fetch side.

Installing it

ComfyUI Manager: search Model Utility Toolkit, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI-ModelUtils

Restart ComfyUI. No downloads needed for the node itself - whether it finds anything useful depends entirely on whether your checkpoint actually has an associated preview image with embedded data.

Where people get burned

The most likely disappointment: a checkpoint with no preview image at all, or one where the author stripped or never embedded workflow metadata into it. In that case workflow_json and metadata_json come back empty, and that's not a bug - there's simply nothing there to extract. This node reads what's present, it doesn't reconstruct anything the checkpoint's author didn't provide.

The other thing worth knowing: workflow_index only helps if a checkpoint actually has multiple associated examples to choose between. If you're expecting to cycle through several preview variants and only ever get index 0's result regardless of what you set, that checkpoint likely only has one example attached in the first place - check metadata_json at a couple of different indices to confirm before assuming the field isn't working.