Sign In

Basic Workflow: FLUX.2-klein I2I v2.0 | 4-in-1 image editing

3

Jan 24, 2026

(Updated: a month ago)

workflows
Basic Workflow: FLUX.2-klein I2I v2.0  |  4-in-1 image editing

Original link: https://sololo.xyz/article/18-basic-workflow-flux2-klein-i2i-v20-4-in-1-image-editing


This workflow supports FLUX.2 klein in both 4B and 9B versions, and I've tested it on ComfyUI 0.10.0.

It's a 4-in-one workflow that includes Single Image Editing, Inpainting with Mask, Multi-Image Reference, and Single Image Outpainting.

Even though it's 4-in-one, it still sticks to the Basic Workflow series'philosophy—simple, fast, and lightweight. You won't need to install any extra custom nodes besides rgthree-comfy.

All example images shown in this post come from my own LoRA samples.

0.png

If you don't need it, you can bypass or just delete the image comparison node on the far right of the workflow.

A few important notes:

  • If you're not using Inpainting with Mask, Multi-Image Reference, or Single Image Outpainting, make sure to keep those nodes in Bypass.

  • Do not enable two or more of these features at the same time, doing so may lead to unexpected results.

Also, a tip about the models: if the 4B version isn't giving you satisfying results, try switching to 9B. But be aware,when I tested the 9B NVFP4 version, I noticed some noticeable color shifts in the outputs, so keep that in mind if you're using this version.

And one last thing: when switching base model versions, don't forget to update the text encoder version too, otherwise ComfyUI will throw errors.

Alright, here's a quick look at how each feature works in practice.

I. Single image editing

When the other three feature nodes are in Bypass, the workflow basically works as standard Single Image Editing. At this point, it's pretty much the same as my previously released Basic Workflow: FLUX.2-klein I2I v1.0. For detailed usage, you can check out my earlier post: Play with FLUX 2 klein | A Simple Image Editing Test.

1_01.png

One small change from the previous v1.0 is that the image loader node has been replaced with “LoadImageOutput”. This node lets you load images from the previous generation, either manually or automatically. It’s super handy for step-by-step, iterative edits.

II. Inpainting with mask

Sometimes, to avoid dirtying the whole body, we need to use a mask.

2_01.png

Once the mask is enabled, just type the effect you want in the prompt and hit run. One thing worth mentioning is that klein is very prompt-friendly—it handles prompts of different lengths and styles really well, and generally follows your instructions closely.

To enable the mask, right-click on Image1 and select “Open in MaskEditor | Image Canvas” from the menu. The mask editor interface is pretty simple, and for our purposes we only use the basic features, so no need for a deep dive. In short, just smear it freely!

And a quick reminder: if you’re no longer using the Inpainting with Mask feature, make sure to set that node to Bypass. The same goes for the other feature nodes.

III. Multi-image reference

In Play with FLUX 2 klein | A Simple Image Editing Test, I showed a few style transfer examples done purely with prompts. Now, you can take it a step further by using reference images along with your prompts.

3_01.png

Multi-image blending has been around since FLUX.1 Kontext, and it’s a super practical feature for tackling many real-world scenarios. I’d say klein has taken this to the next level. Here’s a quick example blending three images:

3_02.png

IV. Single image Outpainting

Outpainting isn’t just about zooming in on the original image—it’s all about revealing more beyond the edges. I first tried this back in the SD1.5 days using ControlNet, but now klein makes it really simple. Just tell it how much extra you want to see, and even if you leave the prompt blank, it still knows what you’re looking for.

4_01.png

You don’t have to, but sometimes putting a little hint in the prompt can make the results even better.

4_02.png

Tip: Instead of extending all sides at once, try doing one or two sides at a time. You’ll usually get more reliable results.

This is the 8th workflow in the Basic Workflow series, and also the most complex one to create so far. Of course, the interface still stays clean and simple.

Hope you have fun experimenting with it, and thanks for checking it out!

3_02_o.png

sololo.xyz

Download HUNDREDS models for FREE

Support me on Ko-fi: https://ko-fi.com/solo_lee

Telegram channel: https://t.me/sololeeai

Model orders: https://sololo.xyz/commission


Thanks so much for your support!

banner.png

3