Krea2 workflow | ti2i 16GB VRAM CivitAI metadata, upscaling, controlnet, FaceDetailer (+indexing elements), inpainting ComfyUI beginner friendly K2 — v2 — v4

This ComfyUI workflow is for making Krea 2 images from text prompts and refining loaded images. It includes paths for ControlNet, face detailing, inpainting, and upscaling, with image preview and saving.

The graph includes prompt-encoding, image-loading, model-loading, sampling, decoding, preview, and image-saving nodes.

You can enter prompt text or load an image; the graph includes nodes to preview and save images.

To use ControlNet, the published instructions call for cloning comfyui-krea2-controlnet into custom_nodes and obtaining the Krea-2 depth-control LoRA.

The workflow lists these model files: bbox/face_yolov8m.pt, depth-control-lora.safetensors, ema_vae_fp16.safetensors, krea2_raw-Q6_K.gguf, and krea2_turbo-Q4_K_M.gguf.

It also lists qwen_image_vae.safetensors, qwen3-vl-4b-heretic_int8_dynamic_convrot.safetensors, qwen3vl_4b_fp8_scaled.safetensors, seedvr2_ema_7b-Q4_K_M.gguf, and Wan2_1_VAE_fp32.safetensors.

The listed ComfyUI add-ons are ComfyUI_Comfyroll_CustomNodes, comfyui_controlnet_aux, comfyui-custom-scripts, comfyui-image-saver, comfyui-impact-pack, comfyui-inpaint-cropandstitch, and rgthree-comfy.

For v4, the author reports using Krea 2 Raw with a Turbo LoRA at weight 0.6. They mention a ratio for other LoRAs but give no numeric value.

The listed files have no download URLs for ema_vae_fp16.safetensors, krea2_turbo-Q4_K_M.gguf, or seedvr2_ema_7b-Q4_K_M.gguf; check how to obtain them if you need those files.

Vorschlag

No model-use license is listed in the page metadata; check the publisher's terms if reuse rights matter.

Vorschlag