FLUX.2 [klein] 4B

FLUX.2 [klein] 4B

Black Forest Labs' small image model that makes images from text and edits them from reference pictures, fast enough for interactive use on a consumer GPU.

Made by black-forest-labs on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: FLUX.2 [klein] 4B is an image model from Black Forest Labs. One model does text to image and image editing, including editing from several reference pictures. It is a rectified flow transformer of about 4 billion parameters, tuned for speed.

The card puts generation in the sub-second range on suitable hardware. The weights are open under Apache 2.0, including for commercial use.

Who it suits: It suits developers and creators who want quick image generation or editing on a machine they own. The card names a graphics card with about 13 GB of VRAM, such as an RTX 3090 or 4070, as enough.

Industry | Who uses it | How they use it | Example
Software | App developers | Build interactive image features into a product | A live preview as a user types a prompt
Design | Visual designers | Generate and revise pictures from references | Editing a scene using two reference images
Marketing | Content teams | Produce illustrations quickly | A header picture from a short brief
Edge and local tools | Engineers | Run image generation on local hardware | An offline image tool on a workstation

Skip it if you want a hosted image service and have no suitable graphics card.

Where to use it:

• Python with the diffusers library
• ComfyUI, which the card lists as supported
• The BFL API, a hosted option the card names
• Windows, macOS or Linux for the file download

Use case | Who | What they get back
Text to image | Creators and developers | A picture made from a written description
Multi-reference editing | Designers | A revised image guided by several reference pictures
Latency-sensitive apps | Product teams | Images in about a second on suitable hardware

How to set it up: You need a few things in place first.

• Python and pip
• A recent diffusers install, taken from its source repository as the card says
• A graphics card with about 13 GB of VRAM
• About 7.8 GB of disk for the BF16 weights, from the record

1. Install or upgrade diffusers as the card describes.
2. Download the model files, or let the script fetch them by name.
3. Run the Python script from the next section.

pip install -U huggingface_hub
hf download black-forest-labs/FLUX.2-klein-4B

How to use it: You write a short Python script. It loads the pipeline class named Flux2KleinPipeline, turns on CPU offloading to save VRAM, and passes a prompt with a size, a step count and a seed. The card's script saves the first image to a file.

import torch
from diffusers import Flux2KleinPipeline

device = "cuda"
dtype = torch.bfloat16

pipe = Flux2KleinPipeline.from_pretrained("black-forest-labs/FLUX.2-klein-4B", torch_dtype=dtype)
pipe.enable_model_cpu_offload()  # save some VRAM by offloading the model to CPU

prompt = "A cat holding a sign that says hello world"
image = pipe(
    prompt=prompt,
    height=1024,
    width=1024,
    guidance_scale=1.0,
    num_inference_steps=4,
    generator=torch.Generator(device=device).manual_seed(0)
).images[0]
image.save("flux-klein.png")

Example: You swap the prompt line for "a red bicycle against a brick wall at sunrise" and run the script again. With the card's settings the model takes 4 steps at 1024 by 1024 pixels, then writes the new picture to flux-klein.png.

For newcomers and experts: Start by running the script as it is and open flux-klein.png. Change only the prompt first. The card warns that text inside images may come out distorted, so check any words in your picture.

The model is a distilled rectified flow transformer, which is why it needs only 4 steps. It trades some of the quality of larger models for speed and a small memory footprint. The card lists limits: it is not a source of facts, it can amplify biases from its training data, and it may miss the prompt. Prompting style strongly affects how closely it follows.

Write a prompt or add reference images
The pipeline runs 4 denoising steps
The model returns an image
You save it to a file

Good to know:

• License: Apache 2.0
• Cost to run: none named for the weights, though the BFL API is a hosted option and the card shows no price
• Data it reaches: it downloads model files from Hugging Face, and the card's script saves images to your disk
• Not for: unlawful use, harming minors, deceptive or fraudulent content, personal data meant to harm someone, harassment, non-consensual intimate images, or automated high-risk decisions

On Hugging Face, 13 November 2026

• Task: image to image
• Library: diffusers
• Parameters: 3.9B
• Weights: about 7.8 GB in BF16; about 1.9 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 435,643
• Likes: 1,116
• License: apache-2.0
• Languages: en
• Quantized versions: 78
• Fine-tunes: 90
• Tags: diffusers, safetensors, text-to-image, image-editing, flux, diffusion-single-file, image-to-image

Links

• Model: https://huggingface.co/black-forest-labs/FLUX.2-klein-4B
• Files: https://huggingface.co/black-forest-labs/FLUX.2-klein-4B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download black-forest-labs/FLUX.2-klein-4B

Then follow the usage in its model card. Nothing in this kit runs by itself.