Easy to use Flux 2 Klein Image Editing Subgraph with up to four image inputs! Useful for single image modifications, combining multiple images into a new image, or disable all images to use as Flux 2 Klein Text-to-Image generator too.
All your image editing workflows in one simple to use Subgraph. With the right prompting you can:
Change styles of your image (realistic, anime, dystopian, cartoon, etc.)
Add, remove, or edit elements in the image (remove unwanted people, change clothes, change hair, add someone to a photo, etc.)
Create new images based on your reference images (3D figurines, model sheets, scene changes, etc.)
Much more.
A Subgraph workflow I made to make using Flux 2 Klein easy for myself, similar to my Qwen Edit 2511. Information / Instructions are included with the workflow.
With the models used this runs well locally on my NVIDIA GeForce RTX 4070 with 12 GB VRAM.
BASIC USAGE
Load your primary image in Load Image 1.
Images 2, 3, and 4 are optional.
Subgraph values:
image#_enable: true if using that image, else false (image ignored).
override_image_size: false to use Image 1 size. true will use the override_width and override_height size, adjusted by aspect_ratio.
aspect_ratio: 'original' for the same aspect as Image 1. 'fixed' to use the override_width/height values. Otherwise use the aspect selected.
The first prompt is Positive and the second is Negative (CFG must NOT be 1 to use Negative prompt).
steps: 4 is the default. 8 is good. The model is pretty fast so you can even use 50 without a long gen time.
Other values are per usual usage.
If you're not familiar with editing you can request changes to items or styles of the image, insert one image element into another, or create whole new images using the characters or items in the reference image(s).
If you disable all four images it will work fine as a text2image model as well.
Description
Original Release. 2026-01-28.
Compatible with Comfy Frontend 1.39.19.
FAQ
Comments (7)
I'm sorry but this workflow is a nightmare. A hundred re-routes and maze-like patterns with nested sub-graphs makes this really difficult to even look at... and moving anything to make the workflow logical or visually sensible and coherent requires deleting dozens of anchor points, one by one.
With recursive nested sub-graphs editing anything is a huge PITA and I simply could not even get started with this. And that's a shame because I would like to accomplish exactly what this workflow purports to do.
Why would you create a literal puzzle out of a normal comfyui workflow like this?
FIVE LAYERS DEEP?!
This is hostile design.
The reroutes are to keep the connections easy to follow and see what is connected where (rather than overlapping each other and going under nodes). Most of the subgraphs are grouping similar tasks that are repeated in different places. It also simplifies the flow visually as well. I am sorry though if this style doesn't work for you. Entirely understandable. But if I had put everything on one level with no reroutes, for me at least it would be an unfollowable mess and actually harder for me to edit. I can count around 100 functional nodes. Can you imagine trying to lay those all out on a single understandable level workflow? It'd be like a program with no function calls. But then I can understand that's probably because I know what I did and where I have stuff. I can imagine coming in cold would feel more like a maze to figure it out. It is a maze of complexity to handle all the possible combinations and connected and unconnected inputs and to try and handle all those things for the user with a simple top level subgraph. Thank you though for the honest feedback as I do want to know where anyone has trouble, or where it helps them. I will keep it in mind and try and keep future ones simpler, but I will probably still use a similar style over all. But I will try and see if I can reduce the depth in the future. Maybe use more groups to explain what each part is doing so you can know where to look more quickly too.
https://civitai.com/images/121991054
Here's what the workflow looks like to me when I enter the main subgraph. The reroute dots make the flow more easy to follow and clearly see what connects. The additional subgraphs mostly group repetitive functions.
Oh, and to get a nice simple workflow version to learn from, in ComfyUI go to Templates on the left, set the Model Filter to Flux.2 Klein, set Runs On to ComfyUI, and look for "Flux.2 [Klein] 4B Distilled: Image Edit". That will give you the base workflow for 1 or 2 image editing. The 4b fp8 model should run on 12GB VRAM. Also feel free to ask questions.
Hey there. I train loras professionally. I create custom workflows from scratch. Running graphs blindly carries a non-zero risk.
For personal use, my graphs can do whatever the fuck I want them to do and look however I want them to look.
But if I share something on the internet, is is intentionally designed for ease-of-use by the lowest common denominator.
100 anchor points shaped liked spirals and recursive nested sub-graphs is literally hostile design.
Anyone who runs a graph blindly is an idiot.
Running your graph without deconstructing it is gambling.
While I appreciate that you are sharing, you should design your work to be shared with the audience in mind, and with security as a priority.
One sub-graph is already abominable. Nesting them 5 layers deep is OBSCENE.
@trains_r_noisy That's fine if you don't like subgraphs. To me it's more readable. To you it's less readable. As for security yes, but you can argue the same thing for custom nodes, maybe even more so, if you don't read the source code.
However I have made a 'flat' version for you. Check it out and see if you like it better or not. I feel it would be too hard to edit for me, but maybe you're more used to working with it laid out like that. Good news is there were a few things I originally used but ended up not using in the final version so I took them out of the flat version.
https://civitai.com/models/2344161?modelVersionId=2719345
@Squirrelly - hey I appreciate your time. I just want to say: I have never had a workflow infuriate me so much. Not even close. Like I said I build custom workflows daily but when I load them from others it is ***necessary*** to inspect each node and see the flow. You added dozens of individual anchor points, making it virtually impossible to actually see where things connect. It's ironic that you claim to have done so in order to facilitate comprehension, because it plainly does the opposite.
There is no accounting for preference and taste, but your workflow is not simply a matter of my personal preference and taste. I assess everything I do based on how others will be able to use it and comprehend it and understand it. I teach people how to train loras. You have taken it upon yourself to share with others and teach others. That is a sort of responsibility that comes with expectations and consequences.
This is not about ME at all. You can keep trying to insist that somehow it's all just some kooky guy on civit bitching about nothing, but I'm giving you honest and practical feedback. I made my comment because others are experiencing the exact same thing, they just don't have time or the will or guts to say something about it.
I generally don't use other people's workflows, I create my own. But with a new model or complex process, it helps to inspect others' graphs that succeed. You made it impossible to inspect your graph. I have never seen a graph with 5 levels of nested subs or dozens of anchor points like that.
I'll check out your flat version. I appreciate you taking the time, and I apologize for the intensity of my comments, but your WF seriously just incensed me in a way that using comfyui should never do.
Thanks.