StormForgeAI Krea 2 WebUI

A lightweight, local WebUI for Krea 2 Turbo, designed for users who want the power of Krea 2 without having to learn or use ComfyUI.

StormForgeAI Krea 2 WebUI provides a clean, straightforward interface focused on image generation rather than node graphs and workflows. Under the hood, the application launches a headless (background) ComfyUI backend that handles model loading, inference, FP8 support, LoRA application, and low-VRAM optimizations automatically. You never need to interact with ComfyUI directly unless you want to troubleshoot something.

Because the backend is powered by ComfyUI, the application uses the standard ComfyUI folder structure for diffusion models, LoRAs, VAEs, text encoders, and other assets. This means you can easily reuse existing ComfyUI model libraries without duplicating files.

Generation time with a LoRA on an RTX 3060 Laptop GPU (6 GB VRAM) is a blazing 36 seconds using the default settings and 40 seconds with LoRA.

Application Screenshot:

⚠️ APP IS IN ACTIVE DEVELOPMENT ⚠️

---

Latest Update: LoRA Support

StormForgeAI Krea 2 WebUI now includes native Krea 2 LoRA support.

The app automatically discovers LoRAs stored inside ComfyUI/models/loras/, including LoRAs organized in subfolders.

From within the WebUI, you can:

• Select a LoRA or choose None to use the base model
• Adjust the LoRA model strength from -2.0 to 2.0
• Refresh the file lists after adding, removing, or replacing LoRAs
• Apply a LoRA without editing workflows or interacting with nodes

The selected LoRA is inserted directly into the dynamically generated ComfyUI API graph using LoraLoaderModelOnly. This applies the LoRA to the diffusion model while leaving the text encoder unchanged, making it suitable for Krea 2 diffusion-model LoRAs.

When None is selected, the application uses the original base-model path without applying a LoRA.

---

Features

• Clean, beginner-friendly interface
• No node graphs or workflow editing required
• Automatic headless ComfyUI backend
• Native Krea 2 Turbo support
• Native LoRA selection and strength control
• Recursive LoRA discovery, including subfolders
• FP8 model loading
• Low-VRAM optimizations for 6 GB GPUs
• Standard ComfyUI folder structure
• Refreshable model and asset lists
• Simple installation and setup
• Separate backend console for troubleshooting

---

Setup

1. Run install.bat.

   This will:

   • Create a Python virtual environment
   • Install PyTorch
   • Install ComfyUI
   • Install all required dependencies for StormForgeAI Krea 2 WebUI

2. Copy your files into the appropriate folders:

   File	Folder
   Krea 2 Turbo FP8 model	ComfyUI/models/diffusion_models/
   Qwen3-VL FP8 text encoder	ComfyUI/models/text_encoders/
   Qwen-Image VAE	ComfyUI/models/vae/
   Krea 2 LoRAs	ComfyUI/models/loras/

3. Run run.bat.

   This automatically:

   • Starts a headless ComfyUI backend using --lowvram
   • Launches StormForgeAI Krea 2 WebUI
   • Opens the WebUI in your default browser

No manual ComfyUI configuration is required.

---

Using LoRAs

Place your Krea 2 LoRA files inside:

ComfyUI/models/loras/

You can also organize them into subfolders:

ComfyUI/
└── models/
    └── loras/
        ├── styles/
        │   └── example_style.safetensors
        ├── characters/
        │   └── example_character.safetensors
        └── example_lora.safetensors

After adding a LoRA:

1. Launch the application or click Refresh File Lists.
2. Select the LoRA from the LoRA dropdown.
3. Adjust its model strength.
4. Generate your image normally.

Select None to generate with the base model only.

LoRA compatibility depends on how the LoRA was trained. Use LoRAs specifically created for Krea 2 or a compatible model architecture.

---

Model Folder Structure

Since the backend uses ComfyUI internally, all assets follow the standard ComfyUI directory layout.

ComfyUI/
└── models/
    ├── diffusion_models/
    ├── text_encoders/
    ├── vae/
    ├── loras/
    ├── clip_vision/
    ├── controlnet/
    └── ...

If you have previously used ComfyUI, you can simply copy or reuse your existing model folders.

---

Notes

• Default settings (steps=8, cfg=1.0, sampler=euler, scheduler=simple) match the recommended settings for Krea 2 Turbo.
• The required timestep shift is already baked into the model configuration, so no additional sampling node is necessary.
• Tiled VAE Decode is enabled by default to reduce peak VRAM usage during decoding. Disable it if your GPU has sufficient memory and you prefer slightly faster decoding.
• If image generation fails, check the separate Krea2 Backend - ComfyUI console window. It contains the complete error messages and traceback, making troubleshooting much easier.
• If PyTorch cannot detect your GPU, edit the --index-url inside install.bat to match your installed CUDA version, then run the installer again.
• Whenever you add, remove, or replace models, LoRAs, VAEs, or other assets, click Refresh File Lists inside the UI or restart the application.

---

Why This Project?

ComfyUI is an incredibly powerful tool, but it can also feel overwhelming if your goal is simply to generate great images.

StormForgeAI Krea 2 WebUI exists to provide a simpler alternative:

• Install.
• Copy your models.
• Launch the app.
• Start creating.

No node graphs.
No workflow editing.
No hunting for missing nodes.

Just a straightforward interface that lets you focus on generating images while ComfyUI quietly handles the heavy lifting in the background.