📥 Load checkpoint SDXL (+Download from CivitAi)

This is the SDXL version of Bjornulf's "load a checkpoint straight from CivitAI" node. Instead of hunting down an SDXL model, downloading it by hand, and dropping it into models/checkpoints, you pick from a curated in-node list of about 100 SDXL checkpoints, and the node downloads whatever you chose (if you don't already have it) before loading it. One node, no file-management detour.

It's the same node as a plain Load Checkpoint, except the dropdown is CivitAI rather than your local folder. Reach for it when you want to A/B a few base models without leaving ComfyUI or babysitting a terminal.

How it works

When the workflow runs, the node looks at the model you picked, checks whether the file exists locally, and if it doesn't, downloads it from CivitAI into ComfyUI/models/checkpoints/Bjornulf_civitAI/sdxl_1.0/. Then it loads the checkpoint the normal way. The download is a one-time cost - after that the file lives on disk and you can keep using this node, or wire a standard Load Checkpoint node to the same file. Nothing magic; it's a convenience wrapper around the CivitAI download API plus a checkpoint loader.

The inputs and outputs that matter

• image - the model picker. It's called image because it renders as a thumbnail gallery of SDXL checkpoints (roughly 101 of them, each showing its CivitAI preview). This is the input you set.
• civitai_token - your CivitAI API token from your account settings. Only needed the first time, to authorize the download; once the file is local you can leave it empty.

The outputs are model, clip, and vae - feed those into your KSampler, CLIP Text Encode, and VAE Decode. You also get name and civitai_url strings if you want to record which checkpoint made an image.

How to install it

Via ComfyUI Manager: search Bjornulf_custom_nodes, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/justUmen/Bjornulf_custom_nodes
pip install -r Bjornulf_custom_nodes/requirements.txt

Restart ComfyUI afterward. The pip install matters here because the CivitAI nodes depend on civitai-py. On Windows portable builds, run pip with the embedded interpreter (.\python.exe -m pip install ...).

Common issues

Downloads need a valid token, and CivitAI models that are gated or early-access can still refuse even with one - that's a CivitAI-side restriction, not a bug in the node. If a run hangs on first use, confirm the token and that the model is publicly downloadable.

SDXL checkpoints are large (usually 6.5GB+), so the first run on any given model will pause while it pulls the file. That's expected; don't kill it thinking it froze. If you're on a metered or ephemeral environment, remember each new model you pick is a fresh multi-gigabyte download.

The pack fixed Windows encoding/link issues with these CivitAI nodes around v0.71–0.75, so stay reasonably current. And note this is one node in a very large, fast-moving pack - if behavior changes after an update, the changelog on the repo is the place to look. If you specifically want Pony or SD1.5 or FLUX base models, use the matching sibling node; this one only lists SDXL.