Krea 2 ComfyUI: Model Downloads, Native Workflow & Style LoRAs

Krea 2 is an open-source image generation model by Krea AI, trained from scratch with a Diffusion Transformer architecture (~12B parameters). It comes in two complementary variants:

Krea 2 Raw — the full base model at 52-step sampling. Designed for diversity and malleability; ideal for fine-tuning and LoRA training.
Krea 2 Turbo — an 8-step distilled checkpoint for fast, high-quality inference. LoRAs trained on Raw apply seamlessly to Turbo.

License: Krea AI Community License

Architecture, VRAM requirements, developer info, license details

Native workflow guide, Subgraph controls, and Cloud usage

Model Downloads

All ComfyUI-repackaged model files are available from Comfy-Org/Krea-2 on Hugging Face and ModelScope mirror (China, faster).

Diffusion Model

Choose the variant that fits your hardware:

File	Precision	Size	Recommended VRAM	Storage Location	Download
`krea2_turbo_fp8_scaled.safetensors`	FP8	~7 GB	16 GB	`models/diffusion_models/`	HF · ModelScope
`krea2_turbo_bf16.safetensors`	BF16	~24 GB	32 GB	`models/diffusion_models/`	HF · ModelScope
`krea2_turbo_nvfp4.safetensors`	NVFP4	~6 GB	12 GB	`models/diffusion_models/`	HF · ModelScope
`krea2_turbo_mxfp8.safetensors`	MXFP8	~7 GB	16 GB	`models/diffusion_models/`	HF · ModelScope
`krea2_turbo_int8_convrot.safetensors`	INT8	~6 GB	12 GB	`models/diffusion_models/`	HF · ModelScope
`krea2_raw_bf16.safetensors`	BF16	~24 GB	32 GB	`models/diffusion_models/`	HF · ModelScope
`krea2_raw_fp8_scaled.safetensors`	FP8	~7 GB	16 GB	`models/diffusion_models/`	HF · ModelScope
`krea2_raw_int8_convrot.safetensors`	INT8	~6 GB	12 GB	`models/diffusion_models/`	HF · ModelScope

Recommended: Start with krea2_turbo_fp8_scaled.safetensors for most setups — good balance of quality and VRAM usage.

Text Encoder

File	Precision	Size	Storage Location	Download
`qwen3vl_4b_fp8_scaled.safetensors`	FP8	~2.4 GB	`models/text_encoders/`	HF · ModelScope
`qwen3vl_4b_bf16.safetensors`	BF16	~8 GB	`models/text_encoders/`	HF · ModelScope

VAE

File	Size	Storage Location	Download
`qwen_image_vae.safetensors`	~300 MB	`models/vae/`	HF · ModelScope

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/
│       └── (style LoRAs, see below)

Native ComfyUI Workflow

Krea 2 is natively supported in ComfyUI. The official workflow uses Subgraph nodes for modular control.

For a full walkthrough of the native workflow — including the ResolutionSelector, prompt enhancement toggle, and LoRA selector — see the official ComfyOrg tutorial.

Quick Start

After installing the model files above and updating ComfyUI to the latest version:

Open ComfyUI
Go to Workflow → Browse Templates → Krea-2 (or search "Krea-2")
Or download the workflow JSON below and drag it onto the canvas

Workflow Controls

The official T2I subgraph exposes the following controls:

Control	Description
Text String (User Prompt)	The prompt describing the image
prompt_enhance	Toggle LLM-powered prompt expansion on/off
Width / Height	Output resolution (via ResolutionSelector, up to 2K)
Seed	Random seed for reproducibility
enable_lora?	Enable or disable style LoRA
LoRA Strength	Strength of the applied style LoRA
LoRA Trigger Word	Auto-populated trigger word for the selected LoRA

Style LoRAs

Krea 2 ships with 9 official style LoRAs. Place the .safetensors files in ComfyUI/models/loras/.

LoRA	Trigger Word	Recommended Strength	Download
`krea2_darkbrush`	monochrome ink wash style	1.0	HF
`krea2_dotmatrix`	dot matrix printing style	0.8	HF
`krea2_kidsdrawing`	children's drawing style	0.9	HF
`krea2_neondrip`	neon drip paint style	0.8	HF
`krea2_rainywindow`	rainy window effect	0.8	HF
`krea2_retroanime`	retro anime style	0.8	HF
`krea2_softwatercolor`	soft watercolor style	0.9	HF
`krea2_sunsetblur`	sunset glow blur style	0.9	HF
`krea2_vintagetarot`	vintage tarot card style	1.0	HF

For additional community-trained style LoRAs, see the Krea 2 LoRA collection on Hugging Face — including 1,500+ styles by ilkerzgi.

Krea 2 on Hugging Face (Comfy-Org) — all repackaged model files
ModelSpec mirror — faster downloads in China
Krea 2 Tech Report
Official GitHub Repository
Krea 2 Model Page — specs, architecture, VRAM requirements