Skip to content
ComfyUI Wikiβ€’
Help Build a Better ComfyUI Knowledge Base Become a Patron

Flux.1 Krea Dev, GGUF, Nunchaku Versions ComfyUI Complete Usage Guide Tutorial

Flux.1 Krea Dev is an advanced text-to-image model developed through collaboration between Black Forest Labs (BFL) and Krea. This is currently the best open-source FLUX model weights, specifically designed for text-to-image generation. This model is now natively available in ComfyUI. For more details, refer to the official ComfyUI blog: FLUX.1 Krea dev Lands on ComfyUI on Day-1

Model Features

  • Unique Aesthetic Style: Enhanced aesthetics, generated images lack the β€œAI look”
  • Natural Details: Generated images have natural, realistic details
  • Exceptional Realism: Provides outstanding realism and image quality
  • Fully Compatible Architecture: Completely compatible architecture design with FLUX.1 [dev]

Model License This model is released under the flux-1-dev-non-commercial-license

When using the workflow files from this tutorial, if you find missing nodes, it may be due to:

  1. Your ComfyUI is not updated: see How to update ComfyUI
  2. Custom nodes are not installed or updated: see How to install custom nodes

All workflows in this tutorial can be found at ComfyUI-Wiki-Workflows. After installing, find the template and search its name in ComfyUI Manager to install.

ComfyUI-Wiki-Workflows

Flux.1 Krea Dev ComfyUI Original Workflow

Since Flux.1 Krea Dev architecture is compatible with Flux.1 Dev, you can directly use the original Flux.1 Dev workflows in ComfyUI

1. Workflow Files

Download the image or JSON below and drag it into ComfyUI to load the corresponding workflow

Flux Krea Dev Workflow

This is the official template workflow in ComfyUI, you can load it directly from ComfyUI templates

Diffusion model

  1. FP8_scaled version suitable for low VRAM users: flux1-krea-dev_fp8_scaled.safetensors
  2. Original weights, suitable for users with sufficient VRAM: flux1-krea-dev.safetensors

If you have used Flux-related workflows before, the following models are the same and don’t need to be downloaded again

Text encoders

VAE

File save locations:

ComfyUI/
β”œβ”€β”€ models/
β”‚   β”œβ”€β”€ diffusion_models/
β”‚   β”‚   └── flux1-krea-dev_fp8_scaled.safetensors or flux1-krea-dev.safetensors
β”‚   β”œβ”€β”€ text_encoders/
β”‚   β”‚   β”œβ”€β”€ clip_l.safetensors
β”‚   β”‚   └── t5xxl_fp16.safetensors or t5xxl_fp8_e4m3fn.safetensors
β”‚   β”œβ”€β”€ vae/
β”‚   β”‚   └── ae.safetensors

3. Step-by-step Check to Run Workflow

For low VRAM users, this model may not run smoothly on your device. You can try the GGUF or Nunchaku versions

Please refer to the image below to ensure all model files are loaded properly

ComfyUI Flux Krea Dev Workflow

  1. Ensure the Load Diffusion Model node has loaded flux1-krea-dev_fp8_scaled.safetensors or flux1-krea-dev.safetensors
    • flux1-krea-dev_fp8_scaled.safetensors recommended for low VRAM users
    • flux1-krea-dev.safetensors if you have sufficient VRAM like 24GB, you can try this version for better quality
  2. Ensure the following models are loaded in the DualCLIPLoader node:
    • clip_name1: t5xxl_fp16.safetensors or t5xxl_fp8_e4m3fn.safetensors
    • clip_name2: clip_l.safetensors
  3. Ensure ae.safetensors is loaded in the Load VAE node
  4. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter to run the workflow

Flux.1 Krea Dev GGUF Version ComfyUI Workflow

Custom node: ComfyUI-GGUF Please refer to the Custom Node Installation Guide section to learn how to install custom nodes

1. Workflow Template

This workflow template can be found in the ComfyUI-Wiki-Workflows plugin. Please update the corresponding plugin to get the latest workflow, or download the latest workflow directly from below

2. Model Download

Since other models are the same as the Fp8 and full versions, you only need to download the GGUF model. Visit QuantStack/FLUX.1-Krea-dev-GGUF to download a version of the model

If you’re not familiar with the corresponding formats, you can judge directly by model file size - usually larger model files require more VRAM

Save to the ComfyUI/models/unet folder

This tutorial uses flux1-krea-dev-Q4_0.gguf

Corresponding model save location

ComfyUI/
β”œβ”€β”€ models/
β”‚   β”œβ”€β”€ unet/
β”‚   β”‚   └── flux1-krea-dev-Q4_0.gguf
β”‚   β”œβ”€β”€ text_encoders/
β”‚   β”‚   β”œβ”€β”€ clip_l.safetensors
β”‚   β”‚   └── t5xxl_fp8_e4m3fn.safetensors
β”‚   β”œβ”€β”€ vae/
β”‚   β”‚   └── ae.safetensors

VRAM Usage Reference

  • Model: flux1-krea-dev-Q4_0.gguf
  • GPU: 4090 24GB
  • Size: 1024*1024
  • VRAM 53%
  • 1st Gen: 40s
  • 2nd Gen: 17~20s

3. Step-by-step Check to Run Workflow

ComfyUI Flux Krea Dev GGUF Workflow

  1. Ensure the Unet Loader (GGUF) node has loaded flux1-krea-dev-Q4_0.gguf
  2. Enter prompts (in English) in CLIP Text Encode(Prompt)
  3. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter to run the workflow

Other model loading is the same as the full version workflow

Flux.1 Krea Dev Nunchaku Version ComfyUI Workflow

Please install and update the custom node: ComfyUI-nunchaku Please refer to the Custom Node Installation Guide section to learn how to install custom nodes

1. Workflow Template

This workflow template can be found in the ComfyUI-Wiki-Workflows plugin. Please update the corresponding plugin to get the latest workflow, or download the latest workflow directly from below

2. Model Download

Since other models are the same as the Fp8 and full versions, you only need to download the Nunchaku-specific model

Nunchaku uses models divided into two versions

Choose the model version based on your graphics card and save to the ComfyUI/models/diffusion_models folder

This tutorial uses svdq-int4_r32-flux.1-krea-dev.safetensors

Corresponding model save location

ComfyUI/
β”œβ”€β”€ models/
β”‚   β”œβ”€β”€ diffusion_models/
β”‚   β”‚   β”œβ”€β”€ svdq-fp4_r32-flux.1-krea-dev.safetensors  #  Blackwell(50 series)
β”‚   β”‚   └── svdq-int4_r32-flux.1-krea-dev.safetensors # Others
β”‚   β”œβ”€β”€ text_encoders/
β”‚   β”‚   β”œβ”€β”€ clip_l.safetensors
β”‚   β”‚   └── t5xxl_fp8_e4m3fn.safetensors
β”‚   β”œβ”€β”€ vae/
β”‚   β”‚   └── ae.safetensors

VRAM Usage Reference

  • Model: svdq-int4_r32-flux.1-krea-dev.safetensors
  • GPU: 4090 24GB
  • Size: 1024*1024
  • VRAM 53%
  • 1st Gen: 46s
  • 2nd Gen: 5s

3. Step-by-step Check to Run Workflow

ComfyUI Flux Krea Dev Nunchaku Workflow

  1. If you find missing nodes after loading the workflow, this is because you haven’t installed Nunchaku’s Wheel. You can disable all other nodes and run only the nodes in the Step-0 group to install the wheel
  2. Ensure the corresponding model is loaded in the Nunchaku Flux DiT Loader node
    • Note the different versions for 50 series graphics cards and other graphics cards
  3. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter to run the workflow

Other model loading is the same as the full version workflow

Flux.1 Krea Personal Usage Experience

  • From current usage, I feel there’s a significant improvement in aesthetics and realism compared to Flux.1 Dev

  • For character representation, there’s no excessive highlighting or oily appearance, showing very good realism

  • Poor generalization, often producing similar results

  • Poor performance in aerial city views with large-scale architectural details

If you’re interested in comparing Flux.1 Dev and Flux.1 Krea Dev, you can visit my Twitter for related comparison outputs: https://x.com/ComfyUIWiki/status/1951261586200600808

All outputs I shared include original ComfyUI workflows