Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-FluxTrainer

ComfyUI Extension: ComfyUI-FluxTrainer

Repo Name

ComfyUI-FluxTrainer

Author
kijai (Account age: 3057 days)
Nodes
View all nodes(1)
Latest Updated
2025-04-02
Github Stars
1.16K

How to Install ComfyUI-FluxTrainer

Install this extension via the ComfyUI Manager by searching for ComfyUI-FluxTrainer
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-FluxTrainer in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

ComfyUI-FluxTrainer Description

ComfyUI-FluxTrainer adds FLUX training to ComfyUI's node-based workflows, enabling LoRA training and comparison of settings, with experimental LyCORIS support and an untested full-model fine-tuning option.

ComfyUI-FluxTrainer Introduction

ComfyUI-FluxTrainer brings experimental FLUX training into ComfyUI. It wraps a slightly modified version of kohya’s training scripts, so you can build a training workflow in the same node-based interface you use to generate images.

The extension supports LoRA training and includes experimental LyCORIS training support. It also offers a full-model fine-tuning option, but the author has not tested that option. If you already use ComfyUI, the main benefit is familiarity: you can work with the models you use for image generation and build workflows to compare training settings.

This is a work in progress. Both the extension and the training code it relies on may change, and the author notes that the implementation may contain mistakes. Treat the provided settings as examples, not as recommended settings for every dataset.

How ComfyUI-FluxTrainer Works

Think of a ComfyUI workflow as a visual recipe. Each node handles part of a task, and connecting nodes tells ComfyUI how to carry it out. ComfyUI-FluxTrainer adds training to that approach: instead of using a workflow only to generate images, you can use one to configure and run a FLUX training job.

The training itself is handled by code adapted from kohya’s scripts. ComfyUI provides the interface for arranging the steps and trying different configurations. For example, you could keep one workflow for a set of training settings and make a second version to compare changes.

The available training approaches differ in scope:

  • LoRA training creates an adaptation for a model rather than fine-tuning the entire main model. This is the approach used by the example workflow.
  • LyCORIS training is another adaptation approach. Its support in this extension is experimental.
  • Full-model fine-tuning trains the main model more directly. The extension includes this option, but the author has not tested it.

ComfyUI-FluxTrainer Features

  • Training inside ComfyUI: Arrange training as a node workflow in an interface you may already know from image generation.
  • Workflows for comparing settings: You can build separate workflows to test configurations. The author cautions that the default settings reflect what worked for the author’s dataset; they are not guaranteed to suit yours.
  • LoRA training: An example LoRA workflow is provided to show how the nodes can be connected. It uses additional nodes from ComfyUI-KJNodes and optional debugging nodes from rgthree-comfy.
  • Experimental LyCORIS training: This feature uses code from LyCORIS. Because it is experimental, expect its behavior or setup to change.
  • Full-model fine-tuning option: Code from kohya’s scripts provides this option, but it is untested in this extension. Do not assume it will behave like the LoRA example.
  • Shared ComfyUI environment: The extension is intended to let you work with the models and Python environment you use for inference. The author reports no incompatibilities in their setup, though that does not guarantee the same result in every environment.

The extension also includes code from prodigy-plus-schedule-free.

ComfyUI-FluxTrainer workflow screenshot

ComfyUI-FluxTrainer Models

ComfyUI-FluxTrainer does not introduce a separate set of models. Instead, you provide FLUX model files appropriate to the type of training you want to run:

Training typeModel requirement stated by the author
LoRAUse a normal fp8 or fp16 version of the main model.
Full-model fine-tuningUse the fp16 version of the main model. This training option is untested in the extension.

For LoRA training, also make sure the VAE is the non-diffusers version. The author points to the FLUX.1-dev ae.safetensors file.

The readme does not specify separate model requirements for LyCORIS training or claim that fp8 and fp16 produce different artistic results. Choose files according to the training approach and the requirements above, rather than assuming that one file format will improve your images.

Troubleshooting ComfyUI-FluxTrainer

Because this extension is experimental, start by checking the workflow and model files when something does not work as expected.

The example LoRA workflow shows missing nodes.

  1. Check whether the missing nodes belong to ComfyUI-KJNodes, which the example workflow uses.
  2. Check whether they are optional debugging nodes from rgthree-comfy.
  3. Review the example workflow to see which nodes are part of the training path and which are for debugging.

A LoRA training workflow appears to have a model-file problem.

  1. Confirm that the main model is a normal fp8 or fp16 version.
  2. Confirm that the VAE is the non-diffusers version; the author links to FLUX.1-dev ae.safetensors.
  3. If you are attempting full-model fine-tuning instead, confirm that the main model is fp16.

The results are not what you expected. The default settings are simply the last settings the author tried successfully with one dataset. They are not universal recommendations. Use separate workflows to compare settings, and consult the kohya training scripts repository for questions about training itself.

Frequently asked questions

  • Is full-model fine-tuning tested? No. The option is present, but the author describes it as untested.
  • Is LyCORIS support stable? It is described as experimental.
  • Are the default settings recommended for my images? Not necessarily. They worked for the author’s dataset and may not suit yours.
  • Where should I ask for training advice? The author asks that training questions be directed to kohya’s repository, rather than to this extension’s repository.

Learn More about ComfyUI-FluxTrainer

  • Example LoRA workflow: Start here to see how the training nodes fit together in ComfyUI. The workflow uses ComfyUI-KJNodes and optional rgthree-comfy debugging nodes.
  • kohya’s training scripts: The underlying training project and the more appropriate place to investigate training-related questions.
  • LyCORIS: Background on the project used for the extension’s experimental LyCORIS support.
  • FLUX.1-dev ae.safetensors: The non-diffusers VAE file linked by the author for LoRA training.

ComfyUI-FluxTrainer Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

ComfyUI-FluxTrainer detailed guide | ComfyUI