🎈 What is Krea2_LiquidMix?

Krea2_LiquidMix is an experimental merge model created to modify the rendering tendencies of Krea2 while preserving its original prompt understanding as much as possible.

It was created to test how much texture, color balance, lighting, glow effects, and overall rendering characteristics can be controlled. Overall, this model is more suitable for illustration-style rendering, and its ability to generate realistic photographic images is weaker than the original model.

🐲 Main Changes from Krea2 Base

Positive Changes

Added more variation in color and lighting

Improved glow effects and effect rendering

Added stronger non-human parts when generating fantasy, furry, and creature-themed characters

Negative Changes

Some body parts or details may break down when generating realistic images

When trying to create cosplay-style images, the output may lose its photorealistic quality

In photorealistic proportions, the eyes may become overly large and the nose may become overly small, resulting in an image that looks closer to anime-style illustration than realistic photography

🧪 Purpose

This model is not intended to reproduce a specific art style, fixed style, or character.

It is a test model created to observe how merging, LoRA integration, and VAE selection affect the rendering behavior of Krea2-based workflows.

🍩 Recommended Use

Comparing different Krea2-based checkpoints

Testing merge behavior

Testing LoRA-integrated Krea2 models

Generating fantasy, furry, insect, and monster characters

Testing glow, effects, highlights, and high-contrast rendering

Comparing color balance and rendering tendencies between different VAEs

⚙️ Recommended Settings

Sampler: ER SDE / Euler a

Steps: 8

CFG: 1

Resolution: 1024–1536

🧬 Requirements

This model is recommended for use with the Qwen3-VL text encoder used in Krea2 workflows.

Recommended text encoder:

qwen3vl_4b_bf16.safetensors

Recommended VAE:

QwenimageVAE_liquid1127

QwenimageVAE_liquid1127 is recommended for Krea2-based workflows because it provides more stable rendering, better color balance, improved highlight expression, and more consistent output than QwenimageVAE_liquid1087.

📛 Notes

This is not an official Krea2 model.

This is an experimental merge model created for testing purposes.

Unexpected outputs or unstable behavior may occur.

Please do not use this model to create depictions of minor, immature-looking, or pre-reproductive furry characters.

Do not use this model for criminal purposes, and create images at your own responsibility.

⭐ Feedback

Feedback and example generations are welcome.

Comparison examples using different VAEs, LoRAs, samplers, or Krea2-based checkpoints would be especially helpful.