Krea 2 Roma

# Krea2 Roma — a Krea 2 workflow with a Midjourney-style look

Krea 2 with the GonzaLomo checkpoint, the UltraReal LoRA and a Midjourney-style LoRA, tuned for a stronger, more stylized finish. Fast: 8 steps, CFG 1.

## Two ways to use it

Nomad Studio (free, open source, local — https://nomadstudio.app)

Download nomad.engine.krea2-roma.zip and drop it into Nomad Studio (Settings › Packages, or the Marketplace view). It becomes an engine you can pick in every Create flow, or call from a prompt with --v roma. Nomad Studio checks which models you are missing and downloads them for you, with each model's licence shown before the download.

ComfyUI

Load krea2_roma_comfyui_api.json and put the models below in the folders shown.

## Resources used (please respect each author's terms)

- qwen3vl_4b_fp8_scaled.safetensors → text_encoders/ (https://huggingface.co/Comfy-Org/Krea-2/resolve/main/text_encoders/qwen3vl_4b_fp8_scaled.safetensors)

- wan_2.1_vae.safetensors → vae/ (https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors)

- GonzaLomo Krea 2 v4.0 (int8) by GBRX — https://civitai.com/models/2761943 (no credit required; you may sell the images; derivatives allowed) → diffusion_models/krea2/gonzalomoKrea2_v40.safetensors

- Midjourney Style v1.0 K2 by zwaetschge — https://civitai.com/models/601666 (credit required; commercial use only through CivitAI's rental program; no derivatives) → loras/krea2/midjourney_8_2_k2_epoch_20.safetensors

- UltraReal - Krea2, Klein9b (KR2_V2) by vizsumit — https://civitai.com/models/2462105 (no credit required; you may sell the images; derivatives allowed) → loras/krea2/ultra_real_krea2_v2.safetensors

Commercial use: the Midjourney-style LoRA allows commercial use only through CivitAI's rental program and requires credit. If you sell images, check its page first — or use Krea2 Monaco, the same setup without that LoRA, which ships with Nomad Studio.

## Settings

8 steps · CFG 1 · euler / normal · Midjourney-style LoRA at 1.0, UltraReal at 0.6 · 1080×1920 by default · works best with natural-language prompts.