Brekel's Custom Nodes for ComfyUI

A collection of custom nodes for ComfyUI designed to enhance and streamline your workflow.
These nodes provide tools for generating, combining, enhancing, and selecting prompts dynamically, plus utility nodes for resolutions, loading/saving images with extra metadata, and loading LoRAs from a folder.

Table of Contents

• 📦 Installation
• 📖 Nodes Overview
  • 📃 Brekel Prompt Enhancer (LLM)
  • 📃 Brekel Auto Prompt Generator
  • 📃 Brekel Prompt Chooser
  • 📐 Brekel Resolution Selector
  • 🖼️ Brekel Load Image (with Filename & Caption)
  • 💾 Brekel Save Image (PNG/JPG)
  • 🎛️ Brekel Lora Loader (Directory)
• 🔒 Viewport Lock (UI extension)
• 📝 Author

📦 Installation

Recommended: ComfyUI Manager

1. Open the ComfyUI Manager.
2. Click on Install Custom Nodes.
3. Search for Brekel or the name of one of the nodes.
4. Click Install on the desired node pack.
5. Restart ComfyUI.

The manager will automatically handle the installation and download any required dependencies for the Prompt Enhancer (LLM) node.

Manual Installation (Git)

1. Navigate to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes/

2. Clone this repository:

   git clone https://github.com/brekel/ComfyUI-Brekel.git

3. Install the required Python packages for the Prompt Enhancer (LLM) node:

   cd ComfyUI-Brekel
   pip install -r requirements.txt

4. Restart ComfyUI.

The other nodes have no dependencies beyond what ComfyUI already ships with.

📃 Brekel Prompt Enhancer (LLM)

This node uses a local Large Language Model (LLM) to creatively rewrite and enhance a simple input prompt into a more detailed and descriptive one.
No need for 3rd party apps or cloud API calls, all from within ComfyUI

Setup Requirements

LLM Models: This node uses LLM models from the ComfyUI/models/LLM/ directory.
The node expects models in the standard Hugging Face format (i.e., the folder should contain config.json, model.safetensors, etc.).

Some LLM model suggestions:

Model name / link	Model Size
Llama-3.2-1B-Instruct	1B
Llama-3.2-3B-Instruct	3B
Llama-3.1-8B-Uncensored	8B
Qwen2-0.5B-Instruct	0.5B
Qwen2-1.5B-Instruct	1.5B

To install a model:

• Open a command prompt in the ComfyUI/models/LLM/ directory.
• On the HuggingFace page click the three dots icon (just left to the Train button) and select Clone Repository.
• Copy the command that starts with: git clone https://.......
• Paste and execute this command in your command prompt.
• You may need to run the git lfs install first to install Large File Support.
• Some models (like Llama) may need you to request access first from the Model card page, approval make take while

System Prompts: The node reads "system prompts" (instructions for the LLM) from .txt files located in ComfyUI/custom_nodes/ComfyUI-Brekel/prompt_enhancer/.
This folder is included with the installation and contains default system prompts for text-to-image and text-to-video. You can edit these or add your own .txt files to create custom LLM behaviors. (the node will automatically add a line to the prompt to ask the LLM to respect the max_length)

Inputs

Parameter	Type	Description
prompt	STRING	The simple, original prompt you want to enhance.
prefix	STRING	Prefix to prepend at the start of the prompt, for example to add your Lora trigger word(s).
model_name	Dropdown	The LLM to use. The list is populated from your ComfyUI/models/LLM/ folder.
quantization	Dropdown	VRAM saving technique. 16-bit is highest quality, while 8-bit and 4-bit use significantly less VRAM at a slight cost to precision.
memory_management	Dropdown	How to handle the model after use: Keep in VRAM (fastest for re-runs), Offload to CPU (saves VRAM), Unload completely (frees all memory).
system_prompt	Dropdown	The instruction template for the LLM, loaded from the prompt_enhancer folder.
max_length	INT	The maximum number of new tokens (words/characters) the LLM can generate.
creativity	FLOAT	Controls the LLM's temperature. 0.0 is deterministic, while higher values (e.g., 0.8) produce more creative and varied outputs.
seed	INT	The seed for the LLM's random generation process. 0 means random.
postfix	STRING	Postfix to append at the end of the prompt, for example to add your Lora trigger word(s).

📃 Brekel Auto Prompt Generator

This node constructs a complete prompt by combining up to four randomly selected lines from different text files with a optional prefix/postfix. It's a powerful tool for creating complex, semi-randomized prompts while still allowing full control by customizing the text file contents.

How to Use

1. This node reads from .txt files located in the ComfyUI/custom_nodes/ComfyUI-Brekel/auto_prompt_generator/ folder, which is included with the installation.
2. Populate this folder with your own .txt files or edit the existing examples. Each file should contain a list of items, one per line (e.g., a file for styles, another for artists, another for lighting types). You can of course add additional text files. (Refresn Node Definitions to update the node with new text file entries)
3. In ComfyUI, use the random_line_file dropdowns to select the files you want to draw from. Selecting "None" will skip that slot.
4. The node will pick one random line from each selected file and combine them with your prefix and postfix using the specified delimiter.
5. Alternatively, set use_static_prompt to "true" to bypass the random generation and use the text from the static_prompt input instead.

Inputs

Parameter	Type	Description
prefix	STRING	Text that will always appear at the beginning of the prompt.
random_line_file1-4	Dropdown	Select a .txt file from the auto_prompt_generator folder. The node will pick one random line from this file.
seed	INT	The seed used for choosing the random lines from the files.
mode	Dropdown	Random Prompt: Selects random lines from the files. Static Prompt: Uses the use_static_prompt.
static_prompt	STRING	A fixed prompt to use when mode is set to Static Prompt.
postfix	STRING	Text that will always appear at the end of the prompt.
delimiter	STRING	The character(s) used to join the different prompt parts (e.g., ", " or "\n" for a newline).
clean_whitespace	Dropdown	true: Removes any leading/trailing whitespace from the final prompt.

📃 Brekel Prompt Chooser

This node allows you to select a text prompt from a folder of .txt files, either randomly or by a specific index.
It's perfect for workflows where you want to iterate through a predefined list prompts.

How to Use

1. This node uses the ComfyUI/custom_nodes/ComfyUI-Brekel/prompt_chooser/ folder, which is included with the installation.
2. Place your .txt files inside this folder. Each file should contain a single prompt.
   • Example: cat.txt could contain "A cat wearing a wizard hat".
   • Example: dog.txt could contain "A dog riding a skateboard".
   • Add as many of your own files and/or delete the example files as needed.
3. In ComfyUI, set the folder_path to this directory (it should be the default).
4. Choose your selection_mode:
   • Random: Uses the seed to pick a random file. A seed of 0 will be different each time.
   • Index: Picks a file based on its alphabetical order in the folder. The index will wrap around if it's larger than the number of files.

Inputs

Parameter	Type	Description
folder_path	STRING	The full path to the folder containing your .txt prompt files.
selection_mode	Dropdown	Random: Selects a file randomly based on the seed. Index: Selects a specific file by its numerical index.
seed	INT	The seed for the random number generator when in Random mode.
file_index	INT	The index of the file to choose (alphabetically sorted) when in Index mode.

📐 Brekel Resolution Selector

A simple dropdown with common resolution presets, so width, height and the number of frames are set in one place and can be re-used throughout your workflow.

How to Use

1. Pick a preset from the resolution dropdown, presets are included for Flux/Qwen, SDXL, SD 1.5 and video (480p / 540p / 720p / 1080p).
2. Use orientation to flip a preset between landscape and portrait, width and height are swapped when needed (square presets are unaffected).
3. Connect width and height to your latent / empty image node, and num_frames to your video node.
4. num_frames steps in increments of 4 with a default of 121, which keeps it WAN compliant (4n+1).

Inputs

Parameter	Type	Description
resolution	Dropdown	The resolution preset to use, grouped by model family (Flux/Qwen, SDXL, SD 1.5, Video).
orientation	Dropdown	Landscape or Portrait, swaps width and height when the preset does not match.
num_frames	INT	Number of frames to generate, primarily for video workflows.

Outputs

Output	Type	Description
width	INT	The width of the selected preset after orientation.
height	INT	The height of the selected preset after orientation.
num_frames	INT	The number of frames, passed through unchanged.

🖼️ Brekel Load Image (with Filename & Caption)

The same as the standard Load Image node, but with two extra outputs: the filename and the caption embedded in the image.

How to Use

• Connect filename to the filename_prefix of a save node to keep the source name of your images through an img2img or upscale workflow, or use it anywhere else a string is accepted.
• Connect caption to a text encoder or prompt input to re-use the caption that was stored in the image, it stays empty when the image has none.
• The caption is read from the PNG text chunks (parameters, Description, caption) and from the JPG/WEBP EXIF fields (UserComment, ImageDescription), which is what the Brekel Save Image node and most other tools write. The workflow/prompt metadata of a regular ComfyUI image is deliberately ignored, that is not a caption.

Inputs

Parameter	Type	Description
image	Dropdown	The image to load from the ComfyUI input folder, with upload button.

Outputs

Output	Type	Description
image	IMAGE	The loaded image.
mask	MASK	The alpha channel as a mask, like the standard node.
filename	STRING	The name of the file without its extension.
caption	STRING	The caption embedded in the image, empty when there is none.

💾 Brekel Save Image (PNG/JPG)

Like the standard Save Image node, but with control over the file format, the numbering and the embedded metadata.

Key Features

• Save as PNG or JPG, with a quality setting for JPG.
• A save toggle that turns the node into a preview node, the images go to the temp folder and the output folder and its numbering are left untouched.
• start_number sets the first number used, so a sequence can start at any offset.
• fill_gaps scans the output folder and re-uses missing numbers, so deleting image_00003_ while image_00004_ exists makes the next save land on _00003_ again, keeping the range continuous.
• Numbering is shared between PNG and JPG (and existing WEBP files), so the two formats never claim the same number.
• By default the workflow/prompt is embedded (drag & drop the PNG onto the canvas to restore it), connecting the optional caption input embeds that text instead. In a JPG the metadata goes into EXIF, and is dropped rather than failing the save when a large workflow does not fit the 64KB limit.
• Outputs the images and the filename that was written, so it can be chained into other nodes.

Inputs

Parameter	Type	Description
images	IMAGE	The images to save.
save	BOOLEAN	save writes to the output folder, preview only acts like a Preview Image node and leaves the output folder untouched.
filename_prefix	STRING	The prefix for the file to save, supports the usual formatting such as %date:yyyy-MM-dd% or %Empty Latent Image.width%.
file_format	Dropdown	png is lossless and can be dragged onto the canvas to restore the workflow, jpg is smaller and stores metadata in EXIF.
quality	INT	JPG quality (1-100), ignored when saving PNG.
start_number	INT	The first number of the sequence, numbers below this are never used.
fill_gaps	BOOLEAN	Re-use missing numbers in the sequence on disk instead of always appending after the highest one.
caption	STRING	Optional, when connected this text is embedded into the image instead of the workflow/prompt.

Outputs

Output	Type	Description
images	IMAGE	The input images, passed through.
filename	STRING	The name of the last file written, without its extension.

🎛️ Brekel Lora Loader (Directory)

Loads a LoRA from any folder on disk by index instead of picking it from the standard dropdown. Because the index is a number widget it gets the fixed / increment / decrement / randomize control, so a batch of queued runs can step through or randomly pick the LoRAs in a folder.

How to Use

1. Set folder_path to the folder holding your LoRAs (leave it empty to use the default ComfyUI loras folder).
2. Files (.safetensors, .ckpt, .pt) are sorted alphabetically, lora_index picks one and wraps around when it is larger than the number of files.
3. Set the control below lora_index to increment or randomize to walk through the folder over multiple runs.
4. Connect LORA_NAME to a save node's filename_prefix or a text input to record which LoRA was used.
5. Setting both strengths to 0 skips loading entirely and passes the model and clip through unchanged.

Inputs

Parameter	Type	Description
model	MODEL	The model to apply the LoRA to.
clip	CLIP	The CLIP to apply the LoRA to.
folder_path	STRING	Folder to load the LoRAs from, empty falls back to the default ComfyUI loras folder.
lora_index	INT	Index of the LoRA in the alphabetically sorted folder, wraps around. Has the increment/randomize control.
strength_model	FLOAT	How strongly to apply the LoRA to the model.
strength_clip	FLOAT	How strongly to apply the LoRA to the CLIP.

Outputs

Output	Type	Description
MODEL	MODEL	The model with the LoRA applied.
CLIP	CLIP	The CLIP with the LoRA applied.
LORA_NAME	STRING	The name of the loaded LoRA without its extension, None when nothing was loaded.

🔒 Viewport Lock (UI extension)

Not a node, this is a small frontend extension that freezes the graph canvas so an accidental drag or scroll no longer moves your carefully arranged workflow. Nodes stay fully usable, only panning and zooming are blocked.

How to Use

1. Click the lock button in the button cluster at the bottom-right of the canvas (next to the fit-view / minimap / link-visibility toggles). A closed padlock means the viewport is locked.
2. Or press Ctrl + Alt + L to toggle it from anywhere.
3. Or use the command palette entry Toggle viewport lock.
4. The state also lives in Settings → Brekel → Viewport → Lock viewport, so it persists between sessions.

Notes

• While locked, allow_dragcanvas is disabled and mouse-wheel zoom is swallowed before LiteGraph sees it.
• Nodes can still be selected, moved and edited. To freeze those too, uncomment the allow_dragnodes line in js/viewport_lock.js.

📝 Author

These nodes were created by Brekel.

• Website: https://brekel.com