V7 update, thank you.

I think I need to reclaim the normal ver and the perpendicular don't have a "better" one, they are just two ways to process the model.

Additionally, I used private LoRA, so it's normal if you get different graph, I used LoRA to get a more obvious result.

Here are V7 results:

There are the old ver description.

This is the updated version of NoobAICyberFix, because the lastest NoobAICyberFix model merges only CyberRealistic V4. (the newest ver is V5.8 until this model publish.)

In short, CyberFix is an anatomy and limb correction (enhancement) version of NoobAI-XL that uses model merging to address issues with anatomical logic and instability in hands and feet. The Standard version is merged using a standard model weighting method, while the Perpendicular version is merged using another method based on perpendicular vector weighting. Consequently, the model incorporates a slight painterly (impasto) and realistic (3D) style, and you may need to use a LoRA to address these related effects.

ALL checkpoints use CyberFix merges methods and Perpendicular merges methods. Use the NoobAI Epsilon-pred 1.1 and V-pred 1.0 and lastest CyberRealistic model.

I will make my effort to release CyberFix that is compatible with the latest version of CyberRealistic.

:)

Note:

From a logical standpoint on model merging, the standard CyberFix model is generally recommended because it minimizes the impact on the original model. (However, some impact is inevitable, as it is a fusion of parameters from a 2D/anime-style model and a 3D/realistic-style model). That said, based on test results and example images, the Perpendicular model can sometimes achieve better results. The parameters of the noise prediction model are not strict, so the outcome is highly dependent on the relevant settings, parameters, and seed.
A LoRA trained on NoobAI is still effective when used with CyberFix. It is recommended to train the LoRA on the original NoobAI model to achieve better results.

compared result (V58 result):

Search AI Models & Addons

very few results