Home
Creators
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

KREA 2 THE MOST KREATIVE NEW TEXT TO IMAGE AI KING WORKFLOW! (Patreon)

Published:
2026-06-24 19:20:44
Edited:
2026-07-02 22:48:06
Imported:
2026-08
Tags:

Downloads

Content

EDIT 03 JULY: If you got an issue with the RBG-SmartSeedVariance node that's ok, a new update made it not work anymore, I updated the workflow below, just download it from this post (it's still the same name so don't worry, just don't mix the old V2 and the new V2)EDIT 27 JUNE: I just updated the workflow to V2, adding a better prompt enhancer node, combo generation with the turbo lora + IdeoKrea lora.You can check the video right here: https://youtu.be/v6KRngGo10UYou can try the IdeoKrea lora here: https://huggingface.co/Aitrepreneur/IdeoKreaUse the V2 models node installer to install the lora/node missing here: https://www.patreon.com/aitrepreneur/posts/161985642Hey everyone! I've created a workflow for KREA 2, the absolute brand NEW MOST KREATIVE (get it?) TEXT TO IMAGE AI KING RIGHT NOW!KREA 2 is really incredible, it is fast, ultra sharp, amazing at prompt following, with insane "kreativity" and it also completely beats models like Z-image turbo while also somehow going toe-to-toe with IDEOGRAM 4, all while having a much better community license and generating images more than 5 TIMES FASTER!

KREA 2 basically combines the speed of a turbo model with the quality, creativity and ease of use of previous models while also being capable of generating image with json prompting like Ideogram 4, which makes it one of the most impressive and versatile open source image models ever. THIS MODEL IS AMAZING!

You can check out the video right here: https://youtu.be/VML57TTsssI

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/161985642

Here the link to the manual install guide: https://pastebin.com/14LPeZ0t

Down below you will also find the KREA2_ULTRA_WORKFLOW-V2.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

Greg

Any plans of implementing a "combo" with another model in this workflow? I really like its creativity but I still think Z-image Combo produced more "realistic" humans. Also, is the Grain slider missing from the workflow? I don't see it in mine.

Aitrepreneur

I could but the way to make the images better would be to use ideogram for the second model, which kinda takes away the whole reason of generating images faster but I might soon do a V2 version of the workflow, I already see things that I can change for better results

Morbid

text to image is awesome but image to image is not working well or giving same results over and over

waseem sabjee

Im messing with image-to-image now. I dont think its meant to work as an image edit. It seems like a way to create a scene based of another scene. could be a nice way of creating key frames or holding character consistency in different scenes. a great way to carry over context

Morbid

Oooh i c i will try that as well

waseem sabjee

I'm still playing around to find the right balance myself. let me know how it goes for you

Aitrepreneur

Image to image is not the same as image edit, it's used to transform a whole image into another one, you'll need to wait for the special edit model for real image editing

Morbid

You were right it only copies the images description but not takes the image of the image

waseem sabjee

I am eagerly awaiting the edit model.

waseem sabjee

V2 is great! i'm also testing it on the bf16 model

GhostJ27F Sims

this is a great wf! I just wanted to let you know that if you don't connect the clip to the lora nodes you are using then a fixed control after generate won't work properly. It will constantly regen from the first pass every time unless you connect the clip connection to the lora nodes. FYI for those that are going to add detailers to this wf ;) EDIT: never mind, has to be a me issue since that didn't work either.

Ken L. Black

Thank you very much for the updated workflow and video!

Aitrepreneur

Not sure I understood the issue and yes the power lora node doesn't need the clip connection to work

J M

Not sure why but I cannot for the life of me get RBG_Smart_Seed_Variance node to install, I try the update via the Manager and reboot but it still says its missing.

Aitrepreneur

what does it say in the command window? dm me the full error log

Drongo

Thanks for the instructions. I’ve now generated some Loras using the AI Toolkit following your guide, but I have to say that they work best with krea’s standard workflow (image_krea2_turbo_t2i). When I load them using your workflow, I find it harder to recognise the people described in the Lora. It works better if I switch off the txt-t-enhancer and the shaper, but it’s still not as perfect as with the standard workflow (which, in turn, has its own weaknesses). And a second question: if I have two people and two different Lora models. How do I tell it that one person has this Lora and the other person follows the other Lora? Even in the text prompt builder – which is designed to allow me to describe different areas differently – it doesn’t respond to my trigger words, but instead turns each person into a mix of both Lora models.

Sean Rowe

Is ther e a good workflow to go from Image to json text prompt builder? Taking an existing image and creating a json prompt out of it

Aitrepreneur

If you are using my workflow, with the person lora, use the normal text to image one, not the combo and don't use t-enhancer or smart seed variance. For the 2 different lora, that's unfortunaly how it works, it has always been an issue, to avoid this you would need to use a regional prompting node workflow and it has always been very hit or miss sometimes, I'll need to look into it myself. Most of the time we would now use special edit models with image reference so lora training wouldn't be as necessary

Aitrepreneur

You can ask chatgpt or use AI-toolkit using the ideogram 4 captioner

Sean Rowe

I'm getting an error now when running this workflow...happened all the sudden. TypeError: krea2t_enhancer_wrapper() takes from 4 to 6 positional arguments but 9 were given https://pastebin.com/LEisjYvU

Aitrepreneur

I can't reproduce the issue so make sure that everything is up to date. 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, redownload the current version of the workflow, redrag and drop it inside comfyui, then try again

Anna

It all seems good, apart from loras, that seem to have no (or no correct effect) at any strength, even though they are working fine in ai-toolkit. I've tried so many ways now, disabling txt options, but my character is never appearing. :(

Aitrepreneur

Make sure that you have updated comfui with the updtae bat file inside the update folder and that you have updated the nodes using the manager inside comfyui first. Then obviously make sure that you have selected the lora with the power lora loader and that you used the same prompt for testing as you did in ai toolkit ( at least at first) you can also make sure you have downloaded a bunch of your loras at different steps and use the xy plot lora workflow to compare them side by side and choose which one is the best

Anna

I guess it may be that for the turbo model you need to train with turbo mode in ai toolkit.

Aitrepreneur

No you certainly don't quite the contrairy, if your lora doesn't work there is a problem with your setup

James Hanggi

Hello my friend! I loaded the KREA model, restarted comfyui and received the following error message: could not load model, node threw an error during execution.

Lucy Moss

I really like upscaled results with combo workflow. Could you create a workflow where we could upload any image and make creative upscaling?

Aitrepreneur

You can already do that with image to image workload just modify the demoise value and choose a higher resolution image for that

Lucy Moss

I tried Image to image upscaling but high resolution like 2160x2704 it completely destroys image quality

AnNyeong

It generates fully blank (all black) images for me

Mark Lonsinger

I have the same problem. Will not install the Smart Seed Variance node.