Krea 2 Support

[Acly]

Krea 2

State: Experimental, basic generation only

Requirements

Requires plugin version 1.52.0 or later.

For custom ComfyUI installs remember to update:

• ComfyUI
• comfyui-tooling-nodes

Model Download

Models need to be downloaded manually. If you are using a plugin-managed installation, please locate the models folder in the place where you installed the server. Download the models to the appropriate sub-folder indicated below.

Overview

📂 models/
 ├── 📂 diffusion_models/
 │   └── krea2_turbo_fp8_scaled.safetensors
 ├── 📂 text_encoders/
 │   └── qwen3vl_4b_fp8_scaled.safetensors
 └── 📂 vae/
     └── qwen_image_vae.safetensors

diffusion_models

• krea2_turbo_fp8_scaled.safetensors
• or one of the other variants from Comfy-Org/Krea-2
• GGUF is not supported

text_encoders

• qwen3vl_4b_fp8_scaled.safetensors

vae

• qwen_image_vae.safetensors

[NoMansPC]

Is this model better than Z Image Turbo for realism, or is it for anime only?

[jmguru]

I like z-image turbo's realism better than Krea 2 so far, but I'm still experimenting. Once we get controlnets, that might change for me. And if we get an edit version of Krea, oof! Good bye z-image.

As are as anime. Krea 2 does it very well in my opinion. I've been playing with these style loras..

[noobie-iv]

Krea can use reference images via custom node:
https://github.com/ethanfel/ComfyUI-Krea2TextEncoder

Is it possible to add this feature as fake controlnet style reference?

[fukc-gihtub]

Also https://huggingface.co/ostris/krea2_turbo_style_reference

[jmguru]

Another thing to consider for Krea 2 support is image variations . For me , it's generating superb images but look generally the same in multiple batches. Varying the prompts help and that's generally the workaround.

[NoMansPC]

Another thing to consider for Krea 2 support is image variations . For me , it's generating superb images but look generally the same in multiple batches. Varying the prompts help and that's generally the workaround.

That's concerning. It's also a problem with Z Image Turbo.

[TheGallant]

VAE is removed?

[Acly]

I think it was always a broken link, fixed now

[iyukar]

Krea 2 seems to be able to understand negative prompts to some extend, the turbo fp8 seems to respond well.
It's not used in many comfy workflows but it would be great if they were activated in krita.

[ralienpp]

The plugin complains about missing files, although they're there and I can generate images with Krea2 directly via ComfyUI. How to troubleshoot this?

The plugin is used with Krita 5.3.3, the logs say:

2024-08-19 15:36:37,449 INFO Extension initialized, Version: 1.22.0, Python: 3.10.7 (tags/v3.10.7:6cc6b13, Sep  5 2022, 14:08:36) [MSC v.1933 64 bit (AMD64)]
2026-08-06 07:27:09,189 INFO Inspecting models at http://127.0.0.1:8765/api/etn/model_info/diffusion_models
2026-08-06 07:27:09,273 INFO GGUF support: node is not installed.
...
2026-08-06 07:27:09,275 INFO Krea 2: missing 1 models

What I have in the directory structure:

.
├── diffusion_models
│   ├── flux-2-klein-4b-fp8.safetensors
│   ├── ideogram4_fp8_scaled.safetensors
│   ├── ideogram4_unconditional_fp8_scaled.safetensors
│   ├── krea2_turbo_fp8_scaled.safetensors
│   ├── krea2_turbo_nvfp4.safetensors
│   ├── qwen-image
│   │   └── qwen_image_fp8_e4m3fn.safetensors
│   ├── qwen-image-2512-Q3_K_M.gguf
│   ├── svdq-int4_r32-flux.1-kontext-dev.part
│   ├── svdq-int4_r32-qwen-image.safetensors
│   ├── z_image_bf16.safetensors
│   └── z_image_turbo_fp8_e4m3fn.safetensors
├── text_encoders
│   ├── clip_l.part
│   ├── qwen_2.5_vl_7b_fp8_scaled.safetensors
│   ├── Qwen3-4B-Q4_K_M.gguf
│   ├── qwen_3_4b.safetensors
│   ├── qwen3vl_4b_fp8_scaled.safetensors
│   ├── qwen3vl_8b_fp8_scaled.safetensors
├── vae
│   ├── ae.safetensors
│   ├── flux2-vae.safetensors
│   ├── flux_vae.safetensors
│   └── qwen_image_vae.safetensors

What the ComfyUI server returns:

{
  "flux-2-klein-4b-fp8.safetensors": {
    "base_model": "flux2",
    "is_inpaint": false,
    "type": "klein-4b"
  },
  "ideogram4_fp8_scaled.safetensors": {
    "base_model": "unknown",
    "is_inpaint": false
  },
  "ideogram4_unconditional_fp8_scaled.safetensors": {
    "base_model": "unknown",
    "is_inpaint": false
  },
  "krea2_turbo_fp8_scaled.safetensors": {
    "base_model": "unknown",
    "is_inpaint": false
  },
  "qwen-image/qwen_image_fp8_e4m3fn.safetensors": {
    "base_model": "qwen-image",
    "is_inpaint": false
  },
  "svdq-int4_r32-qwen-image.safetensors": {
    "base_model": "qwen-image",
    "is_inpaint": false,
    "quant": "svdq"
  },
  "z_image_bf16.safetensors": {
    "base_model": "z-image",
    "is_inpaint": false
  },
  "z_image_turbo_fp8_e4m3fn.safetensors": {
    "base_model": "z-image",
    "is_inpaint": false
  }
}

List of checkpoints:

{
  "RealVisXL_V5.0_fp16.safetensors": {
    "base_model": "sdxl",
    "is_inpaint": false,
    "type": "eps"
  },
  "dreamshaper_8.safetensors": {
    "base_model": "sd15",
    "is_inpaint": false
  },
  "flat2DAnimerge_v45Sharp.safetensors": {
    "base_model": "sd15",
    "is_inpaint": false
  },
  "flux1-dev-fp8.safetensors": {
    "base_model": "flux",
    "is_inpaint": false
  },
  "flux1-schnell-fp8.safetensors": {
    "base_model": "flux-schnell",
    "is_inpaint": false
  },
  "juggernautXL_version6Rundiffusion.safetensors": {
    "base_model": "sdxl",
    "is_inpaint": false,
    "type": "eps"
  },
  "pixelwave_11.safetensors": {
    "base_model": "sdxl",
    "is_inpaint": false,
    "type": "eps"
  },
  "realisticVisionV51_v51VAE.safetensors": {
    "base_model": "sd15",
    "is_inpaint": false
  },
  "serenity_v21Safetensors.safetensors": {
    "base_model": "sd15",
    "is_inpaint": false
  },
  "zavychromaxl_v80.safetensors": {
    "base_model": "sdxl",
    "is_inpaint": false,
    "type": "eps"
  }
}

There is indeed no checkpoint with "krea" in the name, but maybe that's not relevant, given that the files listed in the guide are in the expected places.

I am looking forward to your help!

[fukc-gihtub]

Maybe custom_nodes/comfyui-tooling-nodes is out of date?

[ralienpp]

Thank you, that did the trick. One thing was surprising though, the node manager said I had version nightly [3.3.0], which is also the latest version on Github at the time of this writing. I nevertheless pressed "Try update", restarted ComfyUI - and then it worked.