What is FLUX.2 [klein]?

FLUX.2 [klein] is a small, fast Flux.2-family model that handles both image generation and instruction-based image editing in one model.

Lineup:

  • 9B / 9B Base (FLUX Non-Commercial License; non-commercial use only)
  • 4B / 4B Base (Apache 2.0)

It's a bit confusing, but the one without "Base" in its name is the distilled model. Base takes 20 steps, while Distilled can generate in 4.

There's no big difference in quality, so we'll mostly use the Distilled model for generation.

Recommended Settings

  • Resolution
    • Min: 64×64
    • Max: 4MP (2048×2048)
    • Both width and height must be multiples of 16
  • Number of Reference Images
    • Max: 4

Flux.2 [klein] 9B

Model Download (9B)

📂ComfyUI/
└── 📂models/
    ├── 📂diffusion_models/
    │   ├── flux-2-klein-9b-fp8.safetensors
    │   └── flux-2-klein-base-9b-fp8.safetensors
    ├── 📂text_encoders/
    │   └── qwen_3_8b.safetensors
    └── 📂vae/
        └── flux2-vae.safetensors

text2image Base

flux_2_klein_base_9b_text2image.json

The official workflow uses something called Flux2Scheduler, but it doesn't make much difference, so I use Simple to keep the workflow simple.

text2image Distilled

flux_2_klein_9b_text2image.json

Change only the following parameters:

  • CFG : 1.0
  • steps : 4

Image Editing Base

flux_2_klein_base_9b_image_edit.json

The basic recipe is "input image + instruction prompt."

  • VAE Encode the input image and pass it to ReferenceLatent.

Image Editing Distilled

flux_2_klein_9b_image_edit.json

Image Editing (Multi-Reference) Base

You can also feed in multiple images as references.

flux_2_klein_base_9b_image_edit_multi.json
  • Just chain Reference Image → VAE Encode → ReferenceLatent blocks one after another.
  • 2 or 3 images both work. (The limit is 4.)

Image Editing (Multi-Reference) Distilled

flux_2_klein_9b_image_edit_multi.json

Flux.2 [klein] 4B

Model Download (4B)

📂ComfyUI/
└── 📂models/
    ├── 📂diffusion_models/
    │   ├── flux-2-klein-4b.safetensors
    │   └── flux-2-klein-base-4b.safetensors
    ├── 📂text_encoders/
    │   └── qwen_3_4b.safetensors
    └── 📂vae/
        └── flux2-vae.safetensors

Workflow

The basics are exactly the same as 9B. Just replace the model and text encoder with the 4B ones.

text2image

  • flux_2_klein_base_4b_text2image.json
  • flux_2_klein_4b_text2image.json

Image Editing

  • flux_2_klein_base_4b_image_edit.json
  • flux_2_klein_4b_image_edit.json

Image Editing (Multi-Reference)

  • flux_2_klein_base_4b_image_edit_multi.json
  • flux_2_klein_4b_image_edit_multi.json

Capabilities / Examples

These are just a few examples.

Changing the style or removing objects is only a tiny part of image editing. What counts as "image editing"? Depending on how you answer that, the possibilities are endless. Go ahead and come up with your own uses.

Single Image

WorkflowsSingle Image
flux_2_klein_9b_style_transfer.json

Style Transfer

Reskin this into a watercolor illustration on textured paper.
flux_2_klein_9b_environmental_change.json

Environmental / State Change

Change the time to bright midday.
flux_2_klein_9b_object_swap.json

Object Swap / Addition

Replace the ice bear with an ice duck. Add a hat on the duck with light blue, red, and white colors. Add sneakers on the duck.
flux_2_klein_9b_text_edit.json

Text Editing

Edit the text "WELCOME" to "Flux.2".
flux_2_klein_9b_restore.json

Image Restoration

Restore and colorize this black-and-white photo.
flux_2_klein_9b_pose.json

ControlNet-like (Pose)

This doesn't actually use ControlNet's mechanism. By giving it a stick figure or depth map and asking for a realistic image based on it, you can do ControlNet-style tasks as an image edit.

A office lady sitting on outdoor stairs at dusk, matching the pose from the reference image. Evening ambient light, calm urban atmosphere. She wears a long skirt and a black camisole with frills. Natural, realistic photo
flux_2_klein_9b_outpainting.json

inpainting / outpainting

You're not giving it a mask to fill. You just tell it to "fill in the gray areas naturally."

Outpaint the gray areas to extend the scene naturally
flux_2_klein_9b_collage_refinement.json

Collage Refinement

Turn this into a single realistic underwater ruins scene with two robots: a sleek white mecha and a large rusty moss-covered robot.

Multi-Reference Image Editing

WorkflowsMulti-Reference Image Editing
flux_2_klein_9b_style_transfer_multi.json

Style Transfer

Change image 1 to match the style of image 2.
flux_2_klein_9b_object_swap_multi.json

Object / Person Swap

Replace the person in image 1 with the person from image 2.
flux_2_klein_9b_object_add_multi.json

Add Object

Place the airship from image 2 in the sky of image 1,Make the airship prominent (closer to camera)

References