Checkpoint Loader (Info)

Look at your checkpoints folder. It's full of something_v9final2_real_FINAL.safetensors, isn't it. ComfyUI's stock loader shows you exactly that name and nothing else, and if you're like most people you've loaded a "mystery merge" and had zero idea what base it was actually built on.

EasyCheckpointLoader ("Checkpoint Loader (Info)" in the node menu, category EasyLoaders) is a drop-in replacement for the built-in CheckpointLoaderSimple that hands you two extra outputs: a clean filename string you can plug straight into Save Image, and a text readout of the checkpoint's real metadata.

The key thing to know up front: it is not a different loader. Under the hood it calls the exact same comfy.sd.load_checkpoint_guess_config function the stock node does, so the model it produces is bit-for-bit the model you already get. What it adds is conveniences wrapped around that call. If you're happy with your current workflow you don't need this - but if you save generations and want to find them again, keep reading.

What the extra outputs give you

• filename_str - the model name scrubbed into something filesystem-safe. This is the actual reason to install the pack.
• model_info - a text string with what the checkpoint says about itself.
• model, clip, vae - standard outputs, identical to the stock loader. Wire model into the sampler, clip into CLIP Text Encode, vae into VAE Decode. Nothing changes here.

How the filename magic works

The name_mode dropdown has three settings:

• Formatted (Snake_Case) (default) - strips anything that isn't letters, numbers, dashes or underscores, collapses repeated underscores, and trims the edges. My. Awesome v9_final.safetensors becomes My_Awesome_v9_final.
• Clean (Filename) - just drops the extension, keeps the original name.
• Original - the raw name, no changes.

prefix and postfix let you wrap text around the result (say, artist_ at the front for sorting), and max_chars caps the length - the default of 30 is sensible for long Flux-era names, and the tooltip spells out the one weird bit: 0 means no limit. That's a real footgun; people set 0 expecting "very short" and get the whole name back.

Where people get burned: the naming output only helps if you actually use it. Drop filename_str into Save Image's filename_prefix input and every image comes out with a readable name instead of ComfyUI_00042_.png. You can even bake a subfolder into prefix (portraits/) to organize as you save. One thing to remember from the metadata world: filenames are purely cosmetic - the full workflow still rides inside the PNG regardless of what you call it, so renaming never costs you reproducibility.

Reading what a checkpoint actually is

model_info is a plain string, so wire it into a Show Text (Text Multiline) node to see it. For .safetensors files it pulls the embedded modelspec metadata - title, architecture, resolution, format - plus the model config class ComfyUI picked (that's how you finally learn a mystery merge is Flux under the hood and not SDXL). .ckpt files skip the metadata read, and anything with no embedded info just shows the filename and config class. Not a full forensic report, but enough to stop guessing.

Installing it

This is the same install for every node in the pack - you get both EasyLoaders nodes at once:

cd ComfyUI/custom_nodes
git clone https://github.com/RevengerNick/ComfyUI-Easy-Loaders.git

then restart ComfyUI. Or search "Easy Loaders" in ComfyUI Manager and hit install there. No extra Python dependencies - the pack ships no requirements.txt and only imports torch, safetensors, and ComfyUI core, so there's nothing to pip that you don't already have.

The honest caveat: this is a young, single-author pack (first commit December 2025) with essentially no community track record yet. The checkpoint loader is simple enough that the risk is low - it's one function call plus string handling - but don't expect a debugging community to lean on.

When to reach for it

If you batch-generate and your output/ folder is a wall of ComfyUI_00xxx_.png, this node pays for itself in an afternoon. If you download merges off CivitAI and want to know what architecture they actually are, the model_info output is a nice bonus. If neither describes you, the stock loader does the same job with fewer moving parts - which is fine, because this one is a convenience, not a revolution.