📀 Load Checkpoint

This is SDVN's souped-up replacement for the stock Load Checkpoint node, and it's one of the nicer things in the pack. It does the normal job - load a checkpoint, output model/clip/vae - but it folds in two extras that actually matter: it can download a checkpoint straight from a URL if you don't have it yet, and it ships with a curated dropdown of common models that auto-download on first use. So a workflow you share can say "use this checkpoint" and just work on someone else's machine, no "please download the 6GB file and put it here" instructions attached.

For a beginner that download-on-demand behavior is the headline. On a hosted setup where models live behind external URLs anyway, it's genuinely convenient; on your local box it saves you a trip to the browser.

How it works

The README lays out two modes cleanly:

• Pick from the folder. Leave Download_url empty and it behaves like the normal loader - it loads whatever you select in Ckpt_name from your checkpoints folder.
• Download from a URL. Fill in Download_url and turn Download on, and it fetches the checkpoint (CivitAI or Hugging Face direct links both work, aria2c-accelerated when available), saves it to your checkpoints folder under Ckpt_url_name, and loads that.

On top of that, the Ckpt_name dropdown isn't just your local files - it includes a curated list of popular checkpoints (SDVN's own SDVN-Real, DreamShaper, Counterfeit, Deliberate, a bunch of SDXL models, ~90-odd in total). Pick one that isn't downloaded yet and it pulls it automatically. Handy, opinionated, very SDVN.

The inputs that matter

• Ckpt_name - the model to load, from your folder plus the curated list. Set it to None if you're downloading via URL instead.
• Download - the boolean switch. On + a URL present = download-and-load. Off = just load from the folder.
• Download_url - the direct link to a checkpoint file when you're downloading.
• Ckpt_url_name - what to name the downloaded file (default model.safetensors; give it something readable).

Outputs are the standard set plus one bonus: model, clip, vae, and ckpt_path (the on-disk path as a string, useful for feeding metadata or info nodes). Wire model/clip/vae into your sampler and encoders exactly like the stock loader.

Common issues

Download fails on Windows/macOS. aria2c again - it's not installed by default on those systems, and SDVN's download features need it. Install it manually (the README says so) or downloads silently won't happen.

A curated model won't download. These pull from external hosts; a gated Hugging Face repo, a dead link, or a CivitAI model behind login can fail. When in doubt, grab the file yourself and drop it in ComfyUI/models/checkpoints, then pick it from the dropdown.

I set a URL but it loaded the wrong thing. Check the Download toggle and Ckpt_name. If Download is off, the URL is ignored and it loads whatever Ckpt_name points at. The switch is the deciding factor.

Note on hosted platforms: managed ComfyUI services often expect custom models via external URLs rather than an arbitrary curated auto-download, so the URL mode is the reliable path there - the folder-picker curated list may or may not resolve depending on the host.

Installing SDVN_Comfy_node

ComfyUI Manager: search "SDVN_Comfy_node" → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node

Then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from the ComfyUI root and restart. Install aria2c if you're on Windows or macOS so the download side works. The node lives under 📂 SDVN.