V2
Model geared more towards illustration
V1 - Why this model?

I designed Kreation to overcome the limitations of Kreamania. I could no longer expand that model's graphic and painterly capabilities—which had previously fully satisfied me—without compromising its realism.
However, I was unable to achieve specific effects with Kreamania (such as abstract art, Cubism, Symbolism, or stylized drawings). Even when using LoRAs, the shapes remained rigidly tied to realism (for instance, making a square human head was simply impossible).
Rather than attempting to modify Kreamania, I chose to explore a new path with this model, prioritizing an unconstrained painterly and graphic style.
Kreation is intentionally incapable of producing realistic images; its focus is strictly limited to painterly and artistic styles. This way, both approaches can be developed independently.
I'm currently only distributing the optimized inference format (INT8/ConvRot version) to keep performance high and protect the core workflow. The BF16 master weights won't be released publicly for this version. Thanks for understanding!
Description
Steps : 8 - 10
Sampler : heunpp2 / sgm_uniform
Krea2T Enhancer Advanced preferably disabled
The trigger word is not essential.
FAQ
Comments (12)
Please reconsider posting some more weights. I’m on Apple silicon and convrot does nothing for me. I have to convert to MLX to get anything usable.
I'll provide mxfp8 or nvfp4 format in the coming days.
There's a problem with the model in Forge Neo. Could you please check it out?
Thanks
Hey, I'm really sorry, but I don't use Forge Neo, I work exclusively with ComfyUI, so I won't be able to help. You might want to check with the community that uses that interface, as the issues are often the same.
Thanks for the quick reply, but this is strange because the 1.0 model works fine, but the problem is with version 2.0. The problem is with the int8_convrot version in the 2.0 model.
@tim Use MXFP8 file, it offers better quality and is faster (requires less computation) than Int8_ConRot. I believe I read somewhere that ConvRot sometimes causes issues with Forge Neo.
Unfortunately, the Kreation2.0 MXFP8 model version doesn't work in the Forge Neo either. I have an RTX 5060 Ti with 16 GB of VRAM, and version 1.0 works without a problem.
Are you planning to release the 2.0 model in a different version as well?
By the way, version 1.0 is great.
@tim And the nvfp4 format, can you use it in Forge Neo?
I checked, nvfp4 format in Forge Neo model Krea 2 works for me without any problems
@tim nvfp4 uploaded
I'm having some trouble with this model, I'm getting the error: RuntimeError: Error(s) in loading state_dict for SingleStreamDiT: size mismatch for blocks.0.attn.wq.weight:
But, thanks for the help.
@tim It might be time to switch to ComfyUI to get rid of these kinds of issues.



















