📥👑 Load Lora SD1.5 (+Download from CivitAi)
Apply an SD1.5 LoRA from CivitAI and auto-download it
- model
- clip
- model
- clip
- name
- civitai_url
- trigger_words
Same idea as Bjornulf's CivitAI checkpoint nodes, but for LoRAs. You browse a curated in-node list of SD1.5 LoRAs (a big one - around 600 entries), pick one, and the node downloads it from CivitAI if you don't already have it, then applies it to your model and CLIP. It's a Load LoRA node whose dropdown is CivitAI instead of your local models/loras folder.
The nice bonus over a plain LoRA loader: it also spits out the LoRA's trigger words, so you don't have to go back to the model page to remember what keyword activates it.
How it works
You pass in your model and clip, pick a LoRA, and on run the node checks for the file locally. If it's missing, it downloads it from CivitAI into ComfyUI/models/loras/Bjornulf_civitAI/sd_1.5/, then patches the LoRA into your model and CLIP at the strengths you set. It's the standard LoRA apply operation - the CivitAI part is just how the file gets there. After the first download you can keep using this node or point a normal Load LoRA node at the same file.
The inputs and outputs that matter
image- the LoRA picker, rendered as a thumbnail gallery (roughly 614 SD1.5 LoRAs). This is the main thing you choose.modelandclip- the model and CLIP you're applying the LoRA on top of. Wire these from your checkpoint loader.strength_modelandstrength_clip- how hard the LoRA hits (default 1.0 each; the range technically runs −20 to 20, but you'll almost always live between about 0.5 and 1.2). Turnstrength_modeldown if the LoRA is overpowering the composition.civitai_token- your CivitAI API token, only needed for the download.
Outputs: model and clip (chain these into your KSampler / text encode), plus name, civitai_url, and trigger_words as strings. That last one is genuinely useful - feed it into your prompt so you don't forget the activation keyword.
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-py dependency from requirements is what does the downloading. Windows portable: .\python.exe -m pip install ....
Common issues
Downloads need a valid token, and gated LoRAs can refuse - check the token first if a run hangs. Make sure the LoRA actually matches your base model: an SD1.5 LoRA on an SDXL checkpoint won't error cleanly, it'll just produce garbage. This node's list is SD1.5 only, so pair it with an SD1.5 checkpoint (Bjornulf's SD1.5 model selector, or any).
If a LoRA seems to do nothing, you probably need its trigger word in the prompt - that's what the trigger_words output is for. And if it's overcooking the image, drop strength_model toward 0.6–0.8. Use the SDXL, Pony, or Hunyuan sibling nodes for those bases.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| image | COMBO | 614 options: lora_sd_1.5/1990s_Anime_Style_LoRA_60529.jpeg, lora_sd_1.5/1_mb_LORA_trained_in_5_mins_that_does_the_same_thing_as_2.5_gb_model_but_better_84808.jpeg, lora_sd_1.5/20D黑丝_185355.jpeg, lora_sd_1.5/2B__NieR_Automata__LoRA___YorHA_edition_51748.jpeg, lora_sd_1.5/2D_Pixel_Toolkit__2D像素工具包__3615077.jpeg, lora_sd_1.5/3D_CG_Style__Realistic_508199.jpeg, +608 | |
| model | MODEL | — | |
| clip | CLIP | — | |
| strength_model | FLOAT | 1.00-20–20 | — |
| strength_clip | FLOAT | 1.00-20–20 | — |
| civitai_token | STRING | — |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| clip | CLIP | — |
| name | STRING | — |
| civitai_url | STRING | — |
| trigger_words | STRING | — |