Home Creators Posts Import Register Favorites Logout

Downloads

Content

Hey everyone! I've created a workflow for the new KREA 2 edit upgrade! Because yes, Krea 2 is now capable of editing images just like a real edit model! YES!But I went even further and created my very own special LoRA, custom workflow, and even custom nodes. Which now will allow you to use Krea 2 to inpaint, outpaint and edit insanely good images like a real edit model! THIS IS FANTASTIC!

You can check out the video right here: https://youtu.be/WXsXE8MTXGk

I also created a 1-click installer for the Webui for my Patreon supporters for Local AND Runpod, that you can find right here:  https://www.patreon.com/aitrepreneur/posts/krea-2-new-is-165420652

Down below you will also find the KREA2_ULTRA_WORKFLOW-V3.json for you to use!

As always, supporting me on Patreon allows me to keep creating helpful resources like this for the community. Thank you for your support - now go have some fun😉!

Files

Comments

Super Assetto Corsa

How well do you think this Image Editor compare with your own Flux Klein Edit Workflow? It's that much better?

HealingPaint

Where do I get the Inpaint LORA from?

Pixey Luna

Flux2Klein still the king on editing in open source, is a native "editing" model, instead Krea2 Editing, is just s "toy" is not a native editing model and use loras to do something "similar" to edit. Is pretty fun to use sometime , but if you're serious about editing, stick with flux2Klein only. A new Flux 3 is coming probably this end of year, and will be a game changer , image, editing and video in one model. I hope they will made the open weight model as well, that works on 24gb vram.... at list :D

Super Assetto Corsa

Yes, that what I thought, I've been using Flux Klein for a while now, it's very good, better than Nano Banana 2 and Pro in some cases.

Bob

Krea 2 has more lora and more dynamics than flux in community. but, i feel flux is more stable. whereas, krea can create pretty insane images, i mean great images. Also, like flux or qwen they always give different generated images, for example put a hat, and the hat will always be different kinds depends on the random seed. but krea always generate same or similar hat no matter random seed, unless you describe different in prompt.

Aitrepreneur

Pixey Luna explained it very well, nothing can replace a real edit model, Flux Klein is extremely powerful and versatile because it's been out for months and has a huge array of Loras available. This Krea 2 nodes and loras are workflow are just for fun, it's powerful and can actually gives results that can be even better than Flux Klein but again nothing can beat a real edit model. Once the real Krea 2 edit model gets released however, I think there will be no competition, unless Flux 3 gets the crown before that so we'll see

Gillian

Using inpaint - I create a mask with green 255 set - it shows black on save - but when I run it nothing happens?

Aitrepreneur

Are you sure you selected the brush icon on the left? It's the second icon

Gillian

There is nothing on the left side in my version. on the right side I see mask layer, pain layer,and mask activate layer. the only way I get a green mask is if I change to negative? otherwise it's black.

Aitrepreneur

Have you watched my video? Does your screen looks like mine?

QueenBambi

Wow crazy work. Thanks so much for the workflow. Low key I would love to put in a request for a v2 Anima workflow as its making a bigger splash then krea 2.

Adam Booth

When trying a text to image run I get a bunch of errors, first with the Smart Seed Variance node, so I disabled that and now I have errors with the K Sampler. I assume this is because I am using the convrot model? UPDATE: This K sampler issue is not actually an issue with the K Sampler but this: krea2t_enhancer_wrapper() takes from 4 to 6 positional arguments but 7 were given It is caused by an outdated ComfyUI-Krea2T-Enhance node. An update to this custom node should fix it. The Smart Seed Var issue was because i'm an idiot and assumed all the values were set.

Pixey Luna

you need to update the whole ComfyUI to the last version to use the Krea2 worflow v3. Use the file update_confyui.bat inside your comfy folder.

Gillian

In my mask editor page the left hand side stuff is in dark shadow but if I scan over it the brush appears and then paint gives the green mask and the prompt works.

Timeless1175

I am having some trouble running the inpainting. I have the exact setup as your video and i am using green mask but the area got replaced is only showing part of the picture. for example, if i high light a building door and add prompt to replace the green area with a black bear, the generated image color is completely off and the it is only showing part of a bear. I have no idea why this happens. any chance you know whats wrong? thank you!

Shiba Logic

the inpainting just make the green parts stay green but i can then just take that picture into the psuedo and tell it to do it there and it works

Shiba Logic

is the pseudo locked to censored? even with some of the loras it seems to avoid doing stuff like putting a nano miniskirt on a character as if saying "step aside, no gooning allowed" lol

Waub Amik

None of the 3 new sections are working for me. Krea Pseudo edit throws a "native krea two-reference source patch failed when following your video exactly for editing 1 image . Both out paint and in paint just leaves the Green area and changes nothing.

Christopher Middleton

I'm also getting the "native krea two-reference source patch failed" error, but the other two are working for me. This error is when I try to run pseudo edit

Aitrepreneur

Again seems like you comfyui install/nodes are not updated To solve this, first go inside the ComfyUI_windows_portable\update folder and run the update_comfyu.bat file. Then launch comfyui, click the manager button and click on “update all”. Then once everything is updated click restart then try again

Matthew Tucker

I've installed all of the nodes, and the update_comfyui.bat file, update all, restart - I get a runtime error: shape '[1536, 6144]' is invalid for input of size 830504 on the krea2_turbo_int8_convrot model, no error on the mxfp8 model. Thoughts?

Christopher Middleton

I'm using Comfy-Desktop. Might that be the problem? Thanks for the quick replies btw.

Infinity 33

Im getting a weird texture on the images using the standard T2I - they get removed using the two time combo - but still im not sure why that would be happening - i switched to Euler and still the same - The texture is a uniform "grid" of dots - they get bigger or smaller depending on the resolution of the generated image

Alon Ashkenazi

text to image creates weird color artifacts and not the image described at all.

Essence Cartoon

When I try Pseudo-editing, my outputs have artifacts and checkerboards.

a pretty parrot

i had this when some of my model files were corrupted

zurbaran

instale todo, pero siempre aprece esto: Error detected Resolve before running the workflow 2 nodes — 1 error 1 Missing Node Packs Install missing packs to use this workflow. To install missing nodes, first run pip install -U --pre comfyui-manager in your Python environment to install Node Manager, then restart ComfyUI with the --enable-manager flag.

Essence Cartoon

I posted 2 weeks ago and sent a direct message. Can you please help?

Lopez M

did you solve this? I heard there's a 180 kb Lora that just unlocks the uncensored but not sure how to use it here.

Gunnar Salum

Has anybody tried the flow after most recent ComfyUI updates? I am getting the following error: Circular reference detected while resolving output 1 of node 274 (New Subgraph). This creates an infinite loop in link resolution. UniqueID: [undefined:274[O]1]

Frank B

After a Comfy update this work flow is now broken for me,: Circular reference detected while resolving output 1 of node 274 (New Subgraph). This creates an infinite loop in link resolution. UniqueID: [undefined:274[O]1]

Frank B

Yep me too - its broken, if AItrep wont update or advise, then time to find a new work flow I think

Aitrepreneur

No worries I just got back, didn't have internet for more than 2 weeks. Will check out as soon as I can, just give me a bit of time

Simon Kwan

When I follow your instructions, and right click on 'open in Mask Editor / Image Canvas' there is no brush icon or anything else to select on the left. It shows the green color is selected, but when I paint, the result is either black, white, or negative, as the Mask Blending option is always on and one of these options must be selected. Below that, there is no way to deselect Mask Layer or to select Paint Layer instead.

Gillian

I fixed it somehow. Works fine now.

PulsarDave

Hey Has this been looked into, wondering if you got an update

Aitrepreneur

Update comfyui and all the nodes, the issue is with the group nodes not enabling/disabling correctly because of a bad comfui update

Gunnar Salum (edited)

Comment edits

2026-10-19 12:00 Not sure if others in this thread are experiencing the same thing but what I am seeing is really strange: 1. When i load the workflow into a brand new tab in comfyui, it tells me it cant find the workflow but it is visible and usable 2. If I forget to turn off the text prompt builder (with Margo Robbie), it does generate a clean image (might be a bit scrambled with multiple Margos but that's not the point, the image is fully realized). 3. If I turn off that step and do a very simple prompt of Man with a hat. It does generate the image BUT most of the image is colored blocks and a hazy image of the man itself is in the middle. I have not change any settings on the flow so it should be the out of the box experience that worked fine before. So until now, the flow somewhat works altho the output is a bit weird BUT as soon as I load any loras into that flow e.g. Lenovo for Krea2, i start to get the same error again. And after that there is no helping it. even if I remove all loras from the selection, if I bypass lora node or even remove it completely, or restart Comfy the error keeps coming up. I have to open up a brand new tab and load the flow back into that one and then we are back in step 1. Ran Update all multiple times... I do have this error as well with each launch, but as other flows are working fine for me, i have not turned much attention to it... and even not sure how to resolve it: [DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui/components/button.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
2026-10-14 14:44 Ok, got most of it working, even though I had updates done multiple times. One last note though, it seems for me the RBG Smart Seed is acting up, even with subtle impact, it messes up the output image. If I disable it, output images are fine.

Ok, got most of it working, even though I had updates done multiple times. One last note though, it seems for me the RBG Smart Seed is acting up, even with subtle impact, it messes up the output image. If I disable it, output images are fine.

dragonwanderer

1-click installer creates the comfyui instance that doesn't work with your workflow anymore. It has the problem with all the rgthree nodes. Nothing that I did really worked so I just ran antigravity on the whole folder and told it to fix it. After like 15 min it fixed few things and now it works somehow. Cause of the Issue Outdated Repository Branch: Your rgthree-comfy custom node directory was checked out on a stale branch from November 2023 (dev). That ancient version attempted to monkey-patch execution.recursive_output_delete_if_changed, a function that was refactored out of ComfyUI long ago (causing the execution errors you encountered). Even with those lines commented out, nodes like Fast Groups Muter, Fast Groups Bypasser, and Label did not exist in rgthree-comfy back in 2023, which is why ComfyUI reported 26 missing nodes. Windows Unicode Console Encoding: When switching to the modern branch, rgthree-comfy's startup banner printed a party popper emoji (🎉), which caused a UnicodeEncodeError: 'charmap' codec can't encode character under Windows cp1252 encoding, failing the node import silently or during startup. What Was Done to Fix It Updated rgthree-comfy to main: Switched custom_nodes/rgthree-comfy to the official latest main branch. In modern versions, the old monkey-patching hack has been completely removed by the author, and all modern frontend/backend nodes are included. Fixed Unicode Encoding for Windows: Patched py/log.py and __init__.py in rgthree-comfy to handle console encoding errors safely without crashing. Added @set PYTHONUTF8=1 to run_nvidia_gpu.bat, run_nvidia_gpu_fast_fp16_accumulation.bat, and run_cpu.bat so Python always runs in UTF-8 mode on Windows. Verified Loading: Confirmed all 48 rgthree nodes (backend and web frontend extensions including Fast Groups Muter, Fast Groups Bypasser, Label, Power Lora Loader, Image Comparer) load cleanly in 0.1 seconds without errors.