📥 Load checkpoint FLUX Dev (+Download from CivitAi)

This is the FLUX Dev flavor of Bjornulf's CivitAI checkpoint loader. You pick a FLUX.1 Dev checkpoint or fine-tune from a short curated list inside the node (about 15 entries - PixelWave, Copax, Flux Realistic and friends), and if you don't have the file, it downloads it from CivitAI and loads it. Same job as a normal Load Checkpoint node, but sourced from CivitAI instead of your local folder.

Handy if you're checkpoint-shopping in the FLUX space and don't feel like manually pulling multi-gigabyte files. One caveat up front: the author flags this specific node as a work in progress in the README, so treat it as the roughest of the CivitAI-selector family.

How it works

When the workflow runs, the node checks for the selected checkpoint locally and, if it's missing, downloads it from CivitAI into ComfyUI/models/checkpoints/Bjornulf_civitAI/flux_d/, then loads it. After the first pull the file is just a local checkpoint you can also load with the standard node.

Worth understanding what "checkpoint" means for FLUX here: these are the all-in-one CivitAI FLUX packages. A from-scratch FLUX setup usually wants the diffusion model, two CLIP/T5 text encoders, and the VAE as separate files, which is why the author notes the list still needs manual cleanup around diffusers formats.

The inputs and outputs that matter

• image - the model picker, a thumbnail gallery of FLUX Dev checkpoints (about 15). This is the input you set.
• civitai_token - your CivitAI API token from your account settings. Needed only to authorize the download; leave it blank once the file is local.

Outputs are model, clip, and vae, which wire into your sampler, text encoding, and VAE decode. You also get name and civitai_url strings for logging.

How to install it

ComfyUI Manager: search Bjornulf_custom_nodes, install, restart. Manually:

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

Restart ComfyUI. The CivitAI download path relies on civitai-py from that requirements file. Windows portable: .\python.exe -m pip install ....

Common issues

Two things to expect. First, downloads need a valid token, and gated or early-access FLUX models can refuse - if a first run hangs, check the token. Second, FLUX Dev files are large and the encoders are heavy, so the initial download of any model is a real wait, not a freeze.

Because the author marks this node as WIP (the list may include entries that don't cleanly load as a single checkpoint, e.g. diffusers-format or NF4/GGUF variants), if a specific pick fails to load, try a different entry or fall back to a proper FLUX setup with the diffusion model, CLIP/T5 encoders, and VAE loaded separately. And remember FLUX Dev's license is non-commercial - fine for personal use, something to check before any commercial workflow. This node is FLUX Dev only; use the SDXL, SD1.5, or Pony siblings for those bases.