Krea2_Turbo_base

A clean workflow for Krea 2 Turbo. No prompt-enhancement models, no clutter.

Put the three model files in place, type a prompt, hit Run.

What's inside

- Load Diffusion Model / Load CLIP (type: krea2) / Load VAE -> KSampler (8 steps, cfg 1.0, euler, simple, denoise 1.0)

- No negative prompt. ConditioningZeroOut feeds the negative input (cfg 1.0 ignores it anyway)

- Power Lora Loader (rgthree) is already wired in. Flip one switch on to use a style LoRA.

All switches are off by default, so it runs as plain Krea 2

- Resolution Selector for quick aspect-ratio changes

- Three notes on the canvas: required models and folders / a system prompt for ChatGPT, Gemini

or Claude that writes Krea 2 prompts for you / official style LoRAs and their trigger words

Requirements

- ComfyUI 0.28 or newer (tested on 0.35.1). Older builds don't offer "krea2" in Load CLIP's type

- Custom node: rgthree-comfy (search "rgthree" in ComfyUI Manager), used by Power Lora Loader

- Models from https://huggingface.co/Comfy-Org/Krea-2

krea2_turbo_int8_convrot.safetensors -> models/diffusion_models/ (default here)

qwen3vl_4b_fp8_scaled.safetensors -> models/text_encoders/

qwen_image_vae.safetensors -> models/vae/