Skip to content

lora.md

Latest commit

 

History

History
129 lines (77 loc) · 4.45 KB

lora.md

File metadata and controls

129 lines (77 loc) · 4.45 KB

Using LoRAs

The pipelines in Scope support using one or multiple LoRAs to customize concepts and styles used in generations.

Compatibility

  • The streamdiffusionv2 pipeline is compatible with Wan2.1-T2V-1.3B LoRAs.
  • The longlive pipeline is compatible with Wan2.1-T2V-1.3B LoRAs.
  • The reward-forcing pipeline is compatible with Wan2.1-T2V-1.3B LoRAs.
  • The memflow pipeline is compatible with Wan2.1-T2V-1.3B LoRAs.
  • The krea-realtime-video pipeline is compatible with Wan2.1-T2V-14B LoRAs.

Installing LoRAs

Scope supports using LoRAs that can be downloaded from popular hubs such as HuggingFace or CivitAI.

A few LoRAs that you can start with for streamdiffusionv2, longlive, reward-forcing and memflow:

A few LoRAs that you can start with for krea-realtime-video:

Using the Settings Dialog

The easiest way to install LoRAs is through the Settings dialog:

  1. Click the Settings icon (gear) in the header
  2. Select the LoRAs tab
  3. Paste a LoRA URL from HuggingFace or CivitAI into the input field
  4. Click Install

The LoRA will be downloaded and saved to your LoRA directory automatically. Once installed, you can select it from the LoRA Adapters section in the Settings panel.

CivitAI API Token

CivitAI requires an API token for programmatic downloads. You can configure this in one of two ways:

Option 1: Settings Dialog

  1. Click the Settings icon (gear) in the header
  2. Select the API Keys tab
  3. Enter your CivitAI API token and click Save

Option 2: Environment Variable

export CIVITAI_API_TOKEN=your_civitai_token_here

Note: The environment variable takes precedence over a token stored through the UI.

Get your API key at civitai.com/user/account.

Manual Installation

For manual installation, follow the steps below.

Local

If you are running Scope locally you can simply download the LoRA files to your computer and move them to the proper directory.

HuggingFace

CleanShot 2025-11-21 at 12 21 11

Click the download button and move the file to the ~/.daydream-scope/models/lora folder.

CivitAI

CleanShot 2025-11-21 at 12 20 28

Click the download button and move the file to the ~/.daydream-scope/models/lora folder.

Cloud

If you are running the Scope server on a remote machine in the cloud, then we recommend you programmatically download the LoRA files to the remote machine.

HuggingFace

First, copy the link address for the .safetensors file found under "Files and versions" of the LoRA page:

CleanShot.2025-11-21.at.12.05.48.mp4

Then, navigate to /workspace/models/lora and run:

wget -O <file name> <link address>

An example:

wget -O pixar_10_epochs.safetensors https://huggingface.co/Remade-AI/Pixar/resolve/main/pixar_10_epochs.safetensors?download=true

Tip: On some RunPod templates, wget may not be pre-installed. If so, run apt-get install wget first.

CivitAI

First, get a CivitAI API key which is required for programmatic downloads.

Then, copy the link address from the LoRA page:

CleanShot.2025-11-21.at.12.14.03.mp4

Then, navigate to /workspace/models/lora and run:

wget -O <file name> "<link address>&token=<TOKEN>"

<TOKEN> is your API key.

An example:

wget -O arcane-jinx.safetensors "https://civitai.com/api/download/models/1679582?type=Model&format=SafeTensor&token=<TOKEN>"

Warning

Many LoRAs require a trigger keyword in your prompt to activate. Check the LoRA's model page (on HuggingFace or CivitAI) for the required trigger phrase and include it in your prompt.

Important

Make sure to surround the link address including the token with double quotes!