JK Easy Checkpoint Loader

At its core this is the same node as ComfyUI's built-in Checkpoint Loader Simple - pick a checkpoint, get back a model, clip, and vae to build your pipeline on. The one addition that makes it worth a separate node: it also hands you the checkpoint's filename back out as text, in two flavors. Small feature, but genuinely useful the moment you want to label an output or a watermark with which model actually generated it.

The inputs and outputs that matter

There's exactly one input: ckpt_name, a dropdown populated from whatever .safetensors / .ckpt files you've got in your checkpoints folder.

Five outputs come back: the standard MODEL, CLIP, and VAE - wire these into your samplers and text encoders exactly like you would from any checkpoint loader - plus CKPT_NAME_FULL and CKPT_NAME. The distinction between the two is the obvious one for a filename: CKPT_NAME_FULL is presumably the full name as stored (extension and all, possibly with a subfolder path), while CKPT_NAME is the cleaned-up version. Neither is documented beyond the name itself, so if you need an exact answer for your specific checkpoint's path structure, wire both into a text preview node once and compare.

Where the string outputs actually help

This is the node that makes JKEasyWatermark's watermark_text field genuinely dynamic instead of a hardcoded string: wire CKPT_NAME straight into the watermark text input and every image gets stamped with the model that made it, automatically, no manual editing per checkpoint swap. The same string outputs are handy for a Save Image filename prefix, a debug overlay, or logging which checkpoint a given run used - especially if you're batch-testing several checkpoints in one session and want the output filenames to actually tell them apart.

Installing it

Part of the kostenickj/jk-comfyui-helpers pack - already available on comfy.icu, or cd ComfyUI/custom_nodes && git clone https://github.com/kostenickj/jk-comfyui-helpers.git and restart for self-hosted ComfyUI. This particular node needs nothing beyond what a normal checkpoint loader needs - no extra model downloads, no separate dependency. As with the rest of this pack, it's a small, low-visibility personal toolkit rather than something with a big community following, so the README and the node's own field names are your primary reference.

Common issues

• Dropdown is empty. No checkpoints found in your models folder - same failure mode as the stock Checkpoint Loader, unrelated to this node specifically. Check your models/checkpoints path.
• Not sure which string output you want. If your downstream use (a filename prefix, a watermark) cares about exact formatting, wire up a text preview once and look at what CKPT_NAME_FULL versus CKPT_NAME actually returns for your checkpoint naming scheme before committing to one in a larger pipeline.
• Expecting extra loader features. This node doesn't add LoRA stacking, VAE overrides, or clip skip - it's deliberately just the standard loader plus two text outputs. For anything fancier, look elsewhere in the pack (JKLilContext / JKBigContext handle a "base model vs. LoRA'd model" split, for instance) or to a dedicated loader pack.