Skip to content
Cobalt Pear

LoRAs

A LoRA is a small add-on file that changes a model’s style, character or details. Cobalt Pear takes LoRAs from Civitai and Hugging Face links.

Which way to pick

Section titled “Which way to pick”

When you press + LoRA on a model, you choose one of two ways.

Built into the worker By link
Pick it when you keep using the same LoRAs you are trying many LoRAs
Pictures no wait for downloads, even after the worker restarts slower: the worker downloads each LoRA first, and again each time it restarts after being idle
Changing LoRAs build the worker again (many minutes) paste a new link, use it on the next picture
How many no set number — the limit is build time up to 4 in one picture (can be raised)

In short: settled on a set of LoRAs → Built into the worker. Still testing → By link.

Built into the worker

Section titled “Built into the worker”
  1. Press + LoRA, pick Built into the worker, and paste the links.
  2. The app shows one line to put in your copy’s Dockerfile, with a link that opens it.
  3. RunPod builds the worker again. Nothing to press when it’s done — the next picture tells the app.

Until the new build is ready, pictures come out without the new LoRAs and say so.

For each picture you pick which built-in LoRAs to use from the LoRA button next to Send.

Section titled “By link”
  1. Press + LoRA, pick By link, and paste a Civitai or Hugging Face link.
  2. Use it on the next picture. Nothing to build.

The worker downloads the LoRA before the picture, and again each time it starts after being idle.

More than 4 at once

Section titled “More than 4 at once”

By default one picture takes up to 4 linked LoRAs. To allow 8:

  1. On RunPod, open your endpoint → Manage → Edit Endpoint.
  2. Under Environment Variables, add REMOTE_LORA_MAX with the value 8.
  3. Press Save Endpoint.

When a model shows “No LoRA on this worker”

Section titled “When a model shows “No LoRA on this worker””

That model runs on RunPod’s own worker, which takes only a prompt, size and seed. LoRAs need one of the Cobalt Pear worker repos (ComfyUI or A1111), or Custom on a list model.