Edit SD Checkpoint Information

You've got forty checkpoints on disk and exactly one of them is "the good one," and you can't remember which. This node is the notebook you keep next to each model file. Edit SD Checkpoint Information stores a description, notes, and a recommended VAE in a .json sidecar next to the checkpoint (or diffusion model), with a small UI to read, edit, and save it - and it also works on diffusion-model files in models/diffusion_models/, not just full checkpoints.

Mechanism first, because it's the part that's actually good design: for my_model.safetensors it reads and writes my_model.json in the same folder - the same sidecar convention training uses for captions and this pack's LoRA editor uses for LoRAs (see image-io-metadata.md for the broader pattern). The JSON is dead simple:

{
  "description": "SDXL base model",
  "notes": "Use with sdxl_vae",
  "vae": "sdxl_vae.safetensors"
}

Inputs: model_name (required, a dropdown that lists both checkpoints/ and diffusion_models/), plus optional description (multiline), notes (multiline), and vae (a string naming the recommended VAE, default None). Pick a model and the UI pulls in its sidecar and a thumbnail if one exists next to the file. Outputs mirror the fields: model_name, description, notes, and vae, so you can feed them onward - most usefully, the vae output can drive a VAE loader, giving you "always use this VAE with this model" as data rather than habit.

The genuinely useful case, and why this beats a text file in your Downloads folder:

• Recommended-VAE memory. Some checkpoints want a specific VAE and it's easy to forget which. The vae field is the reminder - and since the node outputs it, you can wire it into a loader and stop carrying that fact in your head entirely.
• Model-swap sanity. When you switch checkpoints in a workflow, the notes field is right there telling you the CFG range or sampler the author recommended. The workflow stays clean; the context stays with the file.
• It's portable, plain JSON. Move the model, move the .json, and the notes follow. No database, no lock-in.

The catches, honestly stated:

• It does not touch the checkpoint itself. Safetensors can embed metadata; this node doesn't read or write that - sidecar only. Delete the .json and the notes are gone.
• The UI is JS-dependent. This is one of the pack's ⚠️ nodes: the read/thumbnail/save interface is a frontend extension that, per the README, may not work under the Nodes 2.0 frontend. The Python output side survives; the buttons and preview may not. Known issue, not your config.

The pack: 2daadv/ComfyUI-GadgetNodes, MIT, one developer, brand new (no community reputation to speak of yet - it's a 2026 release). Install via ComfyUI Manager - search "GadgetNodes" - or git clone https://github.com/2daadv/ComfyUI-GadgetNodes into custom_nodes/, pip install -r requirements.txt, restart. It lives in the Gadget/model menu with the LoRA editor and the pack's unified loader.

If your checkpoint folder is a mess, this is thirty seconds of setup per model that pays back every time you forget which VAE goes with which file - which, if you're like most of us, is every week.