Krea 2 Image-to-Image Depth ControlNet LoRA

Requirements

• GPU: NVIDIA RTX 30XX / 40XX / 50XX (FP8 supported)
• VRAM: 8 GB minimum (CUDA)
• OS: Windows (portable package available)
• Storage: 20 GB+ free

Model downloads

• Krea 2 Turbo GGUF https://huggingface.co/vantagewithai/Krea-2-Turbo-GGUF/tree/main ⧉
• Krea 2 Turbo FP8 https://huggingface.co/AlperKTS/Krea2_FP8/tree/main ⧉
• Qwen3-VL 4B FP8 text encoder https://huggingface.co/Comfy-Org/Qwen3-VL/tree/main/text_encoders ⧉
• Qwen Image VAE https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/vae ⧉
• Depth Control LoRA (862 MB) https://huggingface.co/Patil/Krea-2-depth-controlnet/tree/main ⧉

Custom nodes

• Krea 2 ControlNet node https://github.com/facok/comfyui-krea2-controlnet ⧉
• ComfyUI Manager https://github.com/ltdrdata/ComfyUI-Manager ⧉
• Reference repo (Tanmaypatil123) https://github.com/Tanmaypatil123/Krea-2-controlnet ⧉

Setup steps

1. Clone the custom node — git clone https://github.com/facok/comfyui-krea2-controlnet.git — into ComfyUI/custom_nodes/.
2. Place Krea 2 Turbo GGUF in ComfyUI/models/unet/ and Krea 2 Turbo FP8 in ComfyUI/models/diffusion_models/.
3. Place the Qwen text encoder in ComfyUI/models/text_encoders/ and the Qwen VAE in ComfyUI/models/vae/.
4. Place the depth-control LoRA in ComfyUI/models/loras/.
5. Restart ComfyUI to load the custom node.
6. Load the workflow JSON below, then input an image + text prompt and run.

Notes & tips

• The LoRA is trained on Krea-2-Raw and is compatible with both Raw and Krea-2-Turbo (8-step).
• Works best on photos and renders with clear perspective.
• If nodes are missing, use ComfyUI Manager's "Install missing custom nodes".
• Prefer one-click? A ready-made Krea 2 installer is available.