Skip to main content
Krea 2 is an image generation model from Krea AI, trained from scratch and focused on creative and stylistic exploration. Krea 2 ships as two models designed to work together: train LoRAs on RAW, run inference on Turbo:
  • Krea 2 RAW: the base model with full-step sampling (52 steps). No distillation, diverse and highly malleable, ideal for fine-tuning and LoRA training.
  • Krea 2 Turbo: an 8-step distilled checkpoint for fast, high-quality generation. LoRAs trained on RAW apply seamlessly to Turbo.
License: Krea AI Community License

​
Watch the Launch Event

To learn more about Krea 2, the story behind its development, its capabilities, and what it means for the open-source ecosystem, watch the launch event recording above (featuring Victor Perez, Miguel Lara, and ComfyAnonymous).

​
Available Model Weights

Krea 2 is available in two variants:
  • Krea 2 RAW: the base model with full-step sampling (52 steps). Designed for diversity and malleability, best for fine-tuning and LoRA training.
  • Krea 2 Turbo: an 8-step distilled checkpoint built for fast, high-quality generation. LoRAs trained on RAW apply seamlessly to Turbo.
Related Links:

​
Krea-2 Turbo text-to-image workflow

Krea-2 Turbo text-to-image workflow

Run in Comfy Cloud

Open in Comfy Cloud

Download Workflow

Download JSON or search “Krea-2” in Template Library
Make sure your ComfyUI is updated.Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated. (Desktop version’s update will delay sometime)If nodes are missing when loading a workflow, possible reasons:
  1. You are not using the latest ComfyUI version (Nightly version)
  2. Some nodes failed to import at startup
The workflow is organized into a few parts:
  1. Text to Image (Krea-2 Turbo) subgraph: the core generation pipeline, containing model loading, prompt handling, sampling, and VAE decode
  2. ResolutionSelector: choose your desired output resolution. Krea 2 supports outputs from 1K to 2K — set the megapixels value to 2.0 to get 2K resolution.
  3. CustomCombo (LoRA selector): a pre-built trigger word selector for the available style LoRAs. If you download additional LoRAs, you can customize this selector and pair them with the corresponding LoRA files accordingly.
  4. SaveImage: saves the generated image

Learn about Subgraph

This workflow uses Subgraph nodes for modular processing. Check out the Subgraph documentation to learn how to customize and extend the workflow.

​
Single-click generation

At its simplest, just enter a text prompt in the subgraph, select a resolution, and click Queue Prompt. The defaults (8 steps, prompt enhancement enabled, no LoRA) produce a high-quality image with minimal configuration.

​
Workflow controls

The Text to Image (Krea-2 Turbo) subgraph exposes the following controls:
ControlDescription
Text String (User Prompt)The text prompt describing the image you want to generate
prompt_enhanceToggle LLM-powered prompt expansion on/off
LLM_max_tokenMaximum token length for prompt enhancement
Width / HeightOutput resolution (controlled by ResolutionSelector)
SeedRandom seed for reproducibility
enable_lora?Enable or disable style LoRA
LoRA StrengthStrength of the applied style LoRA
LoRA Trigger WordAuto-populated trigger word for the selected LoRA

​
Style LoRAs

Krea also released a collection of style LoRAs for Krea 2. Select one in the CustomCombo node and enable enable_lora? in the subgraph to apply the style:
LoRATrigger WordRecommended StrengthDownload
krea2_coolblueteal watercolor illustration style0.8Download
krea2_darkbrushmonochrome ink wash style1.0Download
krea2_plasmoidethereal shimmering light style0.8Download
krea2_warmpastelmuted minimalist sketch style0.8Download
Place the .safetensors files in ComfyUI/models/loras/.

​
Model downloads

For local use, download the ComfyUI-optimized model files from Comfy-Org/Krea-2.

Diffusion Model (FP8)

krea2_turbo_fp8_scaled.safetensors: Turbo FP8 (recommended for most users)

Text Encoder (FP8)

qwen3vl_4b_fp8_scaled.safetensors: Qwen3VL-4B text encoder

VAE

qwen_image_vae.safetensors

Style LoRAs

View all style LoRAs on Hugging Face
Other model variants (BF16, NVFP4, MXFP8) are also available for users with higher-end hardware.

​
Model storage location

📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └── krea2_turbo_fp8_scaled.safetensors
│   ├── 📂 text_encoders/
│   │   └── qwen3vl_4b_fp8_scaled.safetensors
│   ├── 📂 vae/
│   │   └── qwen_image_vae.safetensors
│   └── 📂 loras/
│       └── krea2_warmpastel.safetensors (and other style LoRAs)