CivArchive
    Arthemy Comics Krea2 - v1.0
    NSFW
    Preview 135934759
    Preview 135923696
    Preview 135923690
    Preview 135923700
    Preview 135923685
    Preview 135923698
    Preview 135923689
    Preview 135923931
    Preview 135923686
    Preview 135923704
    Preview 135923695
    Preview 135923701
    Preview 135923706
    Preview 135923703
    Preview 135933463

    Arthemy Comics Krea2

    I've baked parts of this filter-bypass in oder to improve the facial expressions: https://civarchive.com/models/2746817/krea2-filter-bypass-fedor

    If you ever worked on a model, you know how it is hard to play with fp8. The process this time has been even messier than usual (so much so that my custom suite for tuning Krea models almost fell apart in any possible way it could), but the result is really fun to work with, if you want a western comics aesthetic or western cartoons.

    As always, I've "trained" this model by manually changing its internal value while checking some live outputs at every step - tuned and equalized its tensors and then built highly specific positive and negative LoRAs trained to solve some of its issues.

    This is really fun to play with but, I want to work a lot more on the TextEncoder side of things - it feels like it's not providing the emotions I'd like to showcase in my images, it's a little stiff, but I think I can solve that... somehow, somewhat, somewhen?

    Let me know if you encounter some bugs or inconsistencies (I'm happy to fix and analyze the errors, but I can't fix what I don't know that it's broken) - and have fun! ^_^

    ___________________________________________________________________________________

    1_____Size: Around 1024 x 1344

    2_____Prompt:

    Quick Prompt

    Western comics style, bold ink outlines, hatched shadows, [expression], seen from [camera angle], upper body portrait, dutch angle, [pose].
    [Character description][clothing][accessories].
    Background: [setting]. Lighting: [light source, color and shadow].

    3_____CFG: 1.0

    4_____Steps: 9

    I hope you like my models and I can't wait to see what you all are going to create with it!

    Description

    FAQ

    Comments (23)

    CosmicSporesJul 6, 2026· 2 reactions
    CivitAI

    Love your stuff! int8-convrot or bf16 coming soon?

    Arthemy
    Author
    Jul 6, 2026· 2 reactions

    For now I just created the fp8 to reflect the original Krea turbo model, but I'm planning on expanding it and I'm not sure which way to take - an fp16 of this model is pretty chunky.

    Carebear1Jul 7, 2026· 2 reactions

    int8-convrot would be great

    Arthemy
    Author
    Jul 7, 2026

    @Carebear1 Unfortunately this model was built natively on the FP8 format - since it was the baseline for Krea-2 turbo (the RAW version is far from precise) It will take me time to convert this model, unfortunately - but if someone manage to convert it from an FP8 I'll be more than happy to add it to the files! ^_^

    Arthemy
    Author
    Jul 12, 2026

    I've posted a bf16 version of the new model, I'll try to convert it to int8 too in the next few days

    Arthemy
    Author
    Jul 13, 2026

    @Carebear1 And int8-convrot has been added.

    wangf35979Jul 7, 2026· 1 reaction
    CivitAI

    An amazing masterpiece!!! I love it so much!

    Arthemy
    Author
    Jul 7, 2026

    Thank you! <3

    SexiamJul 7, 2026· 1 reaction
    CivitAI

    You mentioned you trained a negative Lora. How does that work? Fist time I've heard that term. Amazing checkpoint as always.

    antarekJul 7, 2026· 1 reaction

    Most of Krea 2 checkpoints require a CFG score of 1. Meaning that it disables negative prompt. So to bypass that limitation when using such a low CFG score you rely on a tool that do the same as would a negative prompt if not better, negative scores. So, for exemple, if you want to not have a character with blue eyes you would add a negative score to "blue eyes" and it would prevent generating characters with blue eyes.

    Arthemy
    Author
    Jul 7, 2026· 1 reaction

    Well, with "Negative" LoRA I meant that you can train a LoRA to do what your model is already accomplishing. For example. if the characters have a very static pose, you can train a LoRA on similar static poses (possibly in a style that you also want to avoid) and then apply it with a negative value to the main model in order to "avoid" that aesthetic.

    antarekJul 7, 2026· 1 reaction
    CivitAI

    It’s great to see you working on Krea 2, and I’m eager to test out your creations using such a promising model. As for the issue with emotions, it likely stems from the censorship baked into the model during its training. Fortunately, there are already ways to bypass these restrictions. I’m thinking specifically of the "Krea2 Projector Tuner (BSS)" node, which seems to be the most effective and flexible solution so far.

    Arthemy
    Author
    Jul 7, 2026· 2 reactions

    Ooohh, interesting! I'll try to find a way to increase the expressions without necessarily removing the censorship.

    EXEbeast830Jul 7, 2026· 1 reaction
    CivitAI

    I was away for a while and you already built a v2 Anima and a Krea2? How do they compare?

    Arthemy
    Author
    Jul 7, 2026· 1 reaction

    Krea Is much "wider" in scope and it can create really outstanding images - unfortunately it also seems that the applied censorship of the creators "muted" some concepts (even the character expressions) making it a little weird to use. Anima is a lightweight model, the real successor of Illustrious in my opinion, and it showcases a great expressiveness and amazing colors - unfortunately it feels much more limited when compared with a model four time its size, but it's so fun to use!

    EXEbeast830Jul 7, 2026· 1 reaction

    @Arthemy Thanks a lot for the detailed explanation mate o/

    Ses_AIJul 8, 2026· 2 reactions
    CivitAI

    Awesome model! Will there be standalone krea lora?

    CivAccountJul 9, 2026· 2 reactions

    Please yes, we don't need a whole checkpoint for Krea2 the Loras work so well.

    Arthemy
    Author
    Jul 9, 2026· 2 reactions

    @CivAccount @Ses_AI I've modified 70% of the model, so the resulting LoRA is really slow to distill and might end up being a 7-9GB LoRA.

    Would that be better than a 12GB standalone model?

    Ses_AIJul 9, 2026

    @Arthemy this will probably load forever or wont evel load at all

    Arthemy
    Author
    Jul 10, 2026· 1 reaction

    @Ses_AI Fine-tuned models are not really "LoRA-distillable" since I've worked on the whole model, a LoRA extracted from it would be as much heavy as a full-model T_T

    ronaldmikhailp236Jul 10, 2026· 2 reactions
    CivitAI

    If Arthemy is working on Krea 2, then for me this is the beginning of the Krea 2 era. Stay awesome brother

    Arthemy
    Author
    Jul 10, 2026

    Well, Krea2 is great and is probably the most effective model alongside Ideogram right now, but it's been a true pain to train - on the other hand, Anima was really lightweight and easy to train and shape as much as I wanted, but it's slightly "dumber" and not as much capable as Krea2 when it comes to backgrounds and multiple characters.
    I invite you to test them both!

    Checkpoint
    Krea 2

    Details

    Downloads
    310
    Platform
    CivitAI
    Platform Status
    Available
    Created
    7/6/2026
    Updated
    7/28/2026
    Deleted
    -

    Files