ComfyUI LoRA Sidebar

ComfyUI LoRA Sidebar

A ComfyUI sidebar for browsing your LoRAs as a grid of preview images, keeping your existing sub-folder structure. Plain files next to each LoRA hold the preview and description, no database or JSON.

Features

• Grid view of every LoRA ComfyUI knows about (all configured loras paths)
• Folder navigation with breadcrumbs, mirroring the folders on disk
• Preview images and descriptions picked up from files next to each LoRA; cards show the first line of the description
• Placeholder tile when a LoRA has no preview image
• Filter by name or description (searches the current folder and everything below it)
• Author shown under the name and toggleable author tags to filter the current view, when the file name ends with _[author] (e.g. Painterly_Style_[someone].safetensors)
• Sort by name or by file size (largest or smallest first) from the toolbar's sort dropdown; folders always stay alphabetical and on top. The card tooltip shows the file name and size
• Adjustable card size (saved in ComfyUI settings)
• Hover a card and use its copy button to copy the LoRA in Impact Pack wildcard syntax, e.g. <lora:styles/my_lora:0.6>, my keyword
• Click a card to open a details dialog with the full preview image, keywords, weight and the complete description
• Rename a LoRA (its preview, description and other <name>.* files are renamed with it) and edit its description from the dialog
• Keywords and weight inputs below the description, kept in sync with its keywords: / weight: lines, each with a button to take the value from the fetched Civitai info (trained words / recommended strength)
• Fetch a LoRA's info from Civitai by its sha256 (name, author, base model, type, recommended strength, trained words, tags, description) with copy buttons and a link to the model page. The hash is read from <name>.sha256, or computed and saved there on the first fetch. Fetched info is kept only until ComfyUI restarts; copy what you want to keep into the description
• Show the Civitai sample images of a LoRA in a scrollable bar at the bottom of the details dialog. Hovering an image shows its prompt, negative prompt and generation settings (model, sampler, steps, CFG, seed, size, ...), selectable and with copy buttons, including one that copies everything in A1111's parameters format
• Make a Civitai sample image the LoRA's preview with its Use as preview button (still images, when editing is allowed). The current preview is kept as <name>_old.<ext>

LoRA Mixer node

The LoRA Mixer node (category loaders) picks a random number (between min_count and max_count) of different LoRAs at random from path (a folder within the LoRA directory, empty for its root, optionally including its sub-folders) and outputs one line per LoRA, each with a random weight between min_weight and max_weight and the LoRA's keywords:

<lora:styles/my_lora:0.73>, my_style, retro
<lora:styles/other_lora:0.51>

Change the seed to get a different pick; it's randomized after every run by default. Multi-line keywords are joined with , to keep one line per LoRA. Feed the output into a wildcard or prompt node that understands <lora:...> (e.g. Impact Pack's).

Preview images and descriptions

Put the files next to the LoRA, using the same base name:

loras/
  styles/
    my_lora.safetensors
    my_lora.png           <- preview image
    my_lora.txt           <- description

Image extensions checked, in order: .png, .jpg, .jpeg, .webp, .gif, then the same list as .preview.<ext> (e.g. my_lora.preview.png).

The description is plain text. Two optional lines are picked up for copying:

keywords: my_style, retro
weight: 0.6
Anything else is free-form notes.

The weight is the first number on its line, so a range such as weight: 0.6-0.9 uses 0.6 and Civitai's weight: 1 (0.8 - 1.2) uses 1. Negative weights (-0.5) and a comma as decimal separator (0,7) work too. Without these lines, the whole text is copied as the keywords, except a first line that is only a weight such as 0.7, which is used as the weight. When no weight is given, the LoRA weight used when copying setting is used.

Settings

• LoRA Card Size: card width in the grid (also changed with the zoom buttons)
• LoRA weight used when copying: weight for LoRAs whose description has none
• LoRA sort order: name, size (largest first) or size (smallest first); also set from the toolbar's sort dropdown
• Allow renaming LoRAs and editing descriptions in the LoRA dialog: on by default. Turning it off hides the edit controls, but the extension's rename and save endpoints stay reachable for anyone who can access your ComfyUI server

Installation

Clone into ComfyUI's custom_nodes/ directory and restart ComfyUI:

cd ComfyUI/custom_nodes
git clone https://github.com/alexopus/ComfyUI-LoRA-Sidebar