Repolode
Submit
shootthesound

shootthesound/comfyUI-Realtime-Lora

A ComfyUI custom node pack for training LoRAs from images inside ComfyUI, plus per-block LoRA analysis, selective loading and base-model editing.

View on GitHub
Stars
575
Forks
39
Open issues
15
Last commit
3 months ago

What is it?

ComfyUI Realtime LoRA Toolkit adds nodes to ComfyUI for training LoRAs from your own images without leaving the graph. It wraps three existing training backends: sd-scripts for SDXL and SD 1.5, Musubi Tuner for Z-Image, Qwen Image, Qwen Image Edit, Wan 2.2 and FLUX Klein, and AI-Toolkit for FLUX.1-dev, Z-Image and Wan 2.2. The trained LoRA can be connected straight into an Apply Trained LoRA node.

Beyond training, the pack includes analyzer nodes that score each block of a LoRA by weight magnitude, and selective loaders that let you toggle or scale individual blocks and schedule strength across generation steps. Model layer editors and deep debiaser nodes scale parts of a base model before LoRAs are applied, and save the modified model to disk. Analysis and editing nodes work without installing any training backend.

Key features

  • 01Trains LoRAs for SDXL, SD 1.5, FLUX, Z-Image, Qwen Image, Wan 2.2 and FLUX Klein inside ComfyUI
  • 02Per-block LoRA analyzer with impact-colored selective loaders
  • 03Strength scheduling with 40+ presets across steps
  • 04Model layer editors and sub-component debiasers for base models
  • 05Supports LoRA, LoKR, LoHa and GLoRA formats

Use cases

  • Training a character or face LoRA and testing it immediately in the same workflow
  • Capturing an art style from a handful of reference images
  • Finding which blocks of a LoRA control the subject, style or composition
  • Teaching image editing behaviors with paired source and target images

Quick start

  1. 1

    Clone the repo into your ComfyUI custom_nodes folder

    cd ComfyUI/custom_nodes git clone https://github.com/ShootTheSound/comfyUI-Realtime-Lora
  2. 2

    Restart ComfyUI

Things to know

  • Training requires separate backends (sd-scripts, Musubi Tuner or AI-Toolkit) that must be installed first.
  • Some models need large VRAM or specific bf16/fp16 model downloads for training.
  • RTX 50-series GPUs need a community installer for AI-Toolkit.

In collections

AI image generators & Stable Diffusion tools

Topics

#lora-training#comfyui#stable-diffusion#image-generation#flux#sdxl#custom-nodes#ai#klein#lora

FAQ

What is comfyUI-Realtime-Lora?
A ComfyUI custom node pack for training LoRAs from images inside ComfyUI, plus per-block LoRA analysis, selective loading and base-model editing. ComfyUI Realtime LoRA Toolkit adds nodes to ComfyUI for training LoRAs from your own images without leaving the graph. It wraps three existing training backends: sd-scripts for SDXL and SD 1.5, Musubi Tuner for Z-Image, Qwen Image, Qwen Image Edit, Wan 2.2 and FLUX Klein, and AI-Toolkit for FLUX.1-dev, Z-Image and Wan 2.2. The trained LoRA can be connected straight into an Apply Trained LoRA node.
Is comfyUI-Realtime-Lora free to use? What license does it have?
comfyUI-Realtime-Lora is open source under the MIT license. Check the license terms before using it in commercial products.
Is comfyUI-Realtime-Lora still maintained?
Updated within 6 months. The last commit to comfyUI-Realtime-Lora was on Aug 21, 2026.
What can I use comfyUI-Realtime-Lora for?
Training a character or face LoRA and testing it immediately in the same workflow Capturing an art style from a handful of reference images Finding which blocks of a LoRA control the subject, style or composition Teaching image editing behaviors with paired source and target images
How do I install comfyUI-Realtime-Lora?
Clone the repo into your ComfyUI custom_nodes folder: cd ComfyUI/custom_nodes git clone https://github.com/ShootTheSound/comfyUI-Realtime-Lora Restart ComfyUI
Where can I get comfyUI-Realtime-Lora?
The source code and installation instructions are on GitHub: https://github.com/shootthesound/comfyUI-Realtime-Lora

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects