Krea2 workflow | ti2i 16GB VRAM CivitAI metadata, upscaling, controlnet, FaceDetailer (+indexing elements), inpainting ComfyUI beginner friendly K2 — v2 — v6

Use this Krea 2 v6 ComfyUI workflow to create images from prompts or loaded images, then preview or save them; it also includes paths for ControlNet, inpainting, face detailing, and upscaling.

The graph includes Krea2 control, crop-and-stitch inpainting, face-detailing, and SeedVR2 upscaling nodes.

Prompt text and loaded images enter through `CLIPTextEncode` and `LoadImage`; `PreviewImage` and `Image Saver` provide preview and save outputs.

Setup calls for a diffusion model, text encoder, VAE, and Turbo LoRA.

To use ControlNet, the instructions say to clone `comfyui-krea2-controlnet` into `custom_nodes` and get the linked depth-control LoRA.

ComfyUI node packages named for this workflow: `ComfyUI_Comfyroll_CustomNodes`, `comfyui_controlnet_aux`, `comfyui-custom-scripts`, `comfyui-image-saver`, `comfyui-impact-pack`, `comfyui-inpaint-cropandstitch`, and `rgthree-comfy`.

Named model files include `bbox/face_yolov8m.pt`, `depth-control-lora.safetensors`, `ema_vae_fp16.safetensors`, `krea2_raw-Q6_K.gguf`, `Krea2_TextFusion_Refusal_Reduction.safetensors`, and `krea2_turbo-Q4_K_M.gguf`.

Other named model files are `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`.

Ollama generation and options nodes are included; the instructions name `gemma-4-12B-it-uncensored-heretic.Q8_0.gguf` for Ollama.

Confirm where to get `ema_vae_fp16.safetensors`, `krea2_turbo-Q4_K_M.gguf`, and `seedvr2_ema_7b-Q4_K_M.gguf` before setup; no download URLs are listed for them.

Vorschlag · nicht geprüft