Load Lora From CivitAI
Skip the download page, load by ID
- model
- clip
- MODEL
- CLIP
Instead of the usual routine - open CivitAI, download the .safetensors, drop it in models/loras, restart or refresh, then pick it from a dropdown - this node takes a CivitAI model ID directly and fetches the file itself when the workflow runs. That's the entire pitch, and it's a genuinely useful one if you're running ComfyUI somewhere without a persistent, pre-populated model folder: an API worker, a fresh cloud instance, anywhere you can't count on the file already being there.
What it is and how it works
It's a standard LoRA-apply node underneath - it patches your diffusion model and CLIP text encoder with the LoRA's low-rank weight deltas, same mechanism as any LoRA loader anywhere. The only thing that's different is the source: you give it an ID string, it resolves that against CivitAI at run time, downloads the file, and applies it. There's no local caching described in the README, so treat every run as a fresh fetch unless you know your executor environment persists the download somewhere.
The inputs and outputs that matter
model(MODEL) andclip(CLIP) - required, the pair the LoRA gets patched into.civitai_model_id(STRING) - "The ID of the model to download from CivitAI," per the node's own tooltip. One thing worth flagging clearly: CivitAI's download API is keyed on the model version ID, not the page-level model ID you see in the browser URL bar when you're just looking at a model's overview. If a plain numeric ID from the model's URL doesn't resolve, look for the version-specific ID instead - usually available from the "Copy download link" option on the specific version you want, or visible in the URL once you've selected a version rather than just landed on the model page.strength_modelandstrength_clip(FLOAT, default 1, range -100 to 100, step 0.01) - the usual two dials. Set to 0 to effectively disable the LoRA without unplugging it; negative values invert its effect, which is occasionally useful but rarely what you want on a first try.- Outputs:
MODELandCLIP, wired onward exactly like the outputs of ComfyUI's built-in LoRA loader.
How to install it
- ComfyUI Manager - search "comfyui-model-dynamic-loader," install, restart.
- Manual -
then restart. Nothing extra to install up front - the whole point of the node is that the model file itself isn't fetched until the workflow actually runs.cd ComfyUI/custom_nodes git clone https://github.com/jax-explorer/comfyui-model-dynamic-loader
Common issues & troubleshooting
The ID you copied from an old guide or workflow doesn't resolve anymore. This isn't rare. CivitAI's own archive numbers put roughly one in five older LoRA links as dead over a couple of years - takedowns, license-driven removals, and platform policy changes all eat into the catalog over time, and 2025 specifically saw a lot of that on the model-hosting side. If an ID that used to work now 404s, the model may simply be gone; a community mirror like CivArchive is the standard place people look next.
It resolves on the site but not through this node. CivitAI split its front door in April 2026: civitai.com now serves only SFW content, and the rest of the catalog - anything flagged adult - lives behind civitai.red under a separate Terms of Service. If the model you want is on the adult side and this node's fetch only talks to the standard public download endpoint, it may not have a path to it at all, since that content sits behind its own access rules now. There's no token or auth field on this node to work around that.
Wrong ID format entirely. Model page ID versus model-version ID is the most common mix-up (see above); double-check you're not accidentally passing something else, like a username or a URL slug, into a field that expects a bare numeric ID.
LoRA loads but does nothing visible. Same as any LoRA: check it's actually built for the base checkpoint you're running - CivitAI pages label the base model, and a mismatch is silent rather than an error. Also check whether the model page calls for a trigger word in your prompt; a lot of LoRAs need one to activate their effect at all.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | The diffusion model the LoRA will be applied to. | |
| clip | CLIP | The CLIP model the LoRA will be applied to. | |
| strength_model | FLOAT | 1.00-100–100 | — |
| strength_clip | FLOAT | 1.00-100–100 | — |
| civitai_model_id | STRING | The ID of the model to download from CivitAI. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| CLIP | CLIP | — |