FLUX 3 Klein in ComfyUI: Weights, Workflow and 8 GB INT8

How to run FLUX 3 Klein in ComfyUI

To run FLUX 3 Klein in ComfyUI, accept the license on Hugging Face to get the gated weights, then follow ComfyUI’s FLUX 3 Klein tutorial; ComfyUI says its INT8 version runs on GPUs with 8 GB of VRAM.

Last updated 2026-10-16

Steps

Check your GPU

Black Forest Labs gives an NVIDIA RTX 3090 as the minimum; ComfyUI says its INT8 version runs on 8 GB of VRAM.

Accept the license on Hugging Face

Follow ComfyUI’s FLUX 3 Klein tutorial

Load the workflow and put each file where the tutorial says, including the Qwen3-VL-4B text encoder.

Write detailed prompts

The model card says short prompts give mediocre results; use a prompt upsampler or write longer, specific prompts.

Where to Find the ComfyUI Workflow

ComfyUI maintains a guide for running FLUX 3 Klein (full name FLUX 3 Image Klein), the 5.6-billion-parameter image model from Black Forest Labs. You can find the setup instructions and workflow files in the ComfyUI FLUX 3 Klein tutorial.

Because ComfyUI updates its integration as the model matures, you should always pull the workflow directly from their documentation rather than relying on third-party instructions. The model supports text-to-image generation, instruction-based image editing, combining up to 10 reference images, bounding-box layout control, and text rendering inside images up to 4K resolution.

What You Need Before Starting

Running the model locally in ComfyUI requires a few specific files and hardware checks:

Model Weights: The weights are open, but they are gated. You must log into your account on the Black Forest Labs Hugging Face repository and accept the FLUX Kommunity License v1.0 before you can download the files. For a step-by-step walkthrough of the download process, visit our download guide.
Text Encoder: FLUX 3 Klein is paired with a separate Qwen3-VL-4B text encoder. The ComfyUI tutorial lists the files the workflow loads and where they go.
A Prompt Upsampler: According to the model card, FLUX 3 Klein is meant to be used with a prompt upsampler (prompt expander). Plan to include a prompt expansion step in your preparation.
Compatible Hardware: Your GPU choice determines which version of the model you can load. Read our VRAM and hardware requirements page for full details, or check the summary table below.

Setup Type	Minimum Hardware Claim	Source
Black Forest Labs’ guidance	NVIDIA RTX 3090	Black Forest Labs documentation
INT8 Version	GPU with 8 GB of VRAM	ComfyUI documentation

If you are using an Apple computer, note that as of October 16, 2026, we found no support statement from Black Forest Labs for Mac or Apple Silicon. See our Mac information page for more context.

Common Pitfalls to Avoid

When setting up FLUX 3 Klein in ComfyUI for the first time, users typically run into four main issues:

Using short prompts directly: The model card explicitly notes that feeding short prompts directly into the model gives mediocre results. Always use a prompt upsampler to expand your text descriptions before generating.
Access errors from Hugging Face: The repository is gated. If a download is refused, log in to Hugging Face and accept the license on the repository page first.
Forgetting the text encoder: The 5.6B image model works with the separate Qwen3-VL-4B text encoder. Check that every file the ComfyUI tutorial lists is in place before you run the workflow.
Assuming full commercial rights: The FLUX Kommunity License v1.0 allows non-commercial use, and it lets individuals and companies with under US$5 million in annual revenue run the model in production and use the outputs commercially. However, the model card states this does not by itself allow commercial fine-tuning or redistribution of the model. Because the full LICENSE.md file requires a login and was not reviewed by us, you must read it yourself. You can review the summary on our license overview page.

A Browser Alternative

If you do not have an 8 GB GPU or an RTX 3090, or if you do not want to manage local weights in ComfyUI, you can run the model in your browser on Klein Studio. Klein Studio is an unofficial website and is not affiliated with or endorsed by Black Forest Labs. It runs FLUX 3 Klein through Black Forest Labs’ API.

Every visitor gets 3 free images a day at 1K resolution with no sign-up required. You can generate images from text or upload one image to edit by instruction.

Questions

Can I run FLUX 3 Klein in ComfyUI with 8 GB of VRAM?

Yes, according to ComfyUI's documentation, the INT8 version of FLUX 3 Klein runs on GPUs with 8 GB of VRAM. For the standard unquantized setup, Black Forest Labs' documentation lists an NVIDIA RTX 3090 as the minimum GPU. Read more on our VRAM and hardware requirements page.

Why are my short prompts giving mediocre results in ComfyUI?

The FLUX 3 Klein model card states that the model is meant to be used with a prompt upsampler (also called a prompt expander). Passing short prompts directly to the model gives mediocre results, so you should expand your prompts before generating.

Where do I download the weights for ComfyUI?

The weights are hosted in a gated repository on Hugging Face by Black Forest Labs. You must log into Hugging Face and accept the FLUX Kommunity License v1.0 before downloading the 5.6B image model and the Qwen3-VL-4B text encoder. See our download guide for details.

Does FLUX 3 Klein work with ComfyUI on a Mac?

As of October 16, 2026, we found no statement from Black Forest Labs on Mac or Apple Silicon support for FLUX 3 Klein. You can check our Mac page for any future updates, or use our browser tool instead.