Skip to content
Images

1125-Krea2Turbo-Asian-Utopian

realistic portraits in 8 steps

3 min readupdated

Creator hinablue has released a portrait fine-tune built on the fast Krea 2 Turbo image generator, tailored for natural facial expressions, poses, skin texture, and fabric details. The repository includes two formats out of the box: full diffusion model weights v5.0-R4 (with the portrait adaptation baked in) and a standalone AsianMix v6.0-R4 LoRA (a lightweight add-on file for the original Krea 2 Turbo). Unlike the base model, this release focuses on lifelike Asian portraits with realistic detail and no heavy plastic smoothing. It is a solid fit for illustrators and local ComfyUI users who want fast portraits with natural lighting and texture.

Key features

  • Two installation options: download ready-to-use quantized checkpoints (INT8 ConvRot or the compact NVFP4) or load the lightweight LoRA (223.84 MB, rank 32) on top of base Krea 2 Turbo.
  • Fast generation speed: inherits the Turbo architecture and runs at 8 steps with CFG 1, with no extra speedup LoRAs required.
  • Detailed skin texture: the new v6 LoRA was trained on a 4,420-image dataset with detailed facial feature captions across four regions (Japan, Korea, Taiwan, and Thailand).
  • Ready-to-use ComfyUI workflow: bundles Krea2Turbo with Lora.json, preconfigured with resolution selection nodes and post-processing for film grain, bloom, and sharpening.

What you can do

  • Generate studio and street portraits: use the asian woman trigger phrase along with the templates in caption_templates.txt to create shots with natural daylight and soft shadows.
  • Save VRAM without sacrificing style: grab the 7.98 GB NVFP4 build instead of the 13 GB INT8 ConvRot checkpoint if your GPU and ComfyUI loader support the format.
  • Fine-tune stylization strength: apply the standalone LoRA to base Krea-2-Turbo to dial its weight up or down manually, or combine the full R4 weights with other compatible LoRAs.

Example

Starter prompt from the model card for a window-lit portrait with natural skin texture (ComfyUI settings: 8 steps, CFG 1, euler / simple sampler):
Asian woman, an adult Taiwanese woman with soft oval facial features, natural makeup and airy curtain bangs, wearing a cream linen blouse, seated beside a window in a quiet cafe, soft natural daylight, realistic skin texture with subtle imperfections, relaxed expression, portrait photography, gentle shadows and natural colors.

Read moreDetails
File sizes: 13.00 GiB (INT8 ConvRot), 7.98 GiB (NVFP4), and 223.84 MiB (standalone LoRA rank 32 / alpha 32); actual VRAM usage is higher due to the qwen3vl_4b_fp8_scaled.safetensors text encoder and qwen_image_vae.safetensors VAE.
Training and merge: LoRA v6 was trained for 2 epochs (15,320 steps) in Krea2 Trainer on 4,420 examples, then merged with AsianMix v5.5-X and baked into the v5.0-R4 checkpoint.
Limitations: weights are not directly compatible with Diffusers from_pretrained() and require an up-to-date ComfyUI installation; the Krea 2 Community license restricts commercial use by companies with $1M or more in annual revenue.
Examples
Discuss Telegram 1 103

Will it run?

Runs in INT8

Fits without quantisation.

  • INT8recommendedweights 13 GB · ready files16 GB
  • 4-bitweights 7.3 GB9.5 GB

Tip: go 8-bit.INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: INT8 — from RTX 5080

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    #!/usr/bin/env bash
    # 1125-Krea2Turbo-Asian-Utopian
    set -e
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    # CUDA 13+ → int8_convrot (Comfy-Org's advice), else fp8_scaled
    CU=$(nvidia-smi | grep -oP 'CUDA Version: \K[0-9]+' || echo 12)
    [ "$CU" -ge 13 ] && TORCH=cu130 || TORCH=cu128
    [ -d ~/ComfyUI ] || git clone https://github.com/comfyanonymous/ComfyUI ~/ComfyUI
    cd ~/ComfyUI
    uv venv --python 3.12 --allow-existing && source .venv/bin/activate
    uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/$TORCH
    uv pip install -r requirements.txt "huggingface_hub[cli]"
    # Base files: Krea 2, 8-bit
    if [ "$CU" -ge 13 ]; then
      hf download Comfy-Org/Krea-2 text_encoders/qwen3vl_4b_int8_convrot.safetensors vae/qwen_image_vae.safetensors --local-dir models
    else
      hf download Comfy-Org/Krea-2 text_encoders/qwen3vl_4b_fp8_scaled.safetensors vae/qwen_image_vae.safetensors --local-dir models
    fi
    # Main model: 1125-Krea2Turbo-Asian-Utopian
    hf download hinablue/1125-Krea2Turbo-Asian-Utopian 1125_hinaKrea2Turbo_AsianUtopian_v5.0-R4-int8-convrot.safetensors --local-dir models/diffusion_models
    # Start ComfyUI. From your computer: ssh -L 8188:localhost:8188 user@IP, then http://localhost:8188
    python main.py --listen 127.0.0.1 --port 8188
  3. 3. Use it

    From your computer: ssh -L 8188:localhost:8188 user@machine-IP, then open http://localhost:8188. Or open that address in the browser on the machine's desktop. The model's workflow is usually in its Hugging Face card.

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • SageAttention 3v2.2.0from October 28, 2025

    even faster on RTX 50 / Blackwell

    FP4 attention for Blackwell. Faster than SageAttention 2, but the authors suggest v2 when accuracy matters.

    How to enable: Build from the sageattention3_blackwell folder of the repo; in ComfyUI via KJNodes.

    Earlier generations (2)
    • SageAttention 2++ — Quantised (INT8/FP8) attention with no visible quality loss. The main speed-up for image and video generation.
    • SageAttention 1 — First version: INT8 for Q·K. Still works, but 2++ is faster and more accurate.
  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (3)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
    • xFormers — The older way to speed up attention, works almost everywhere. Sage or Flash are faster where available.
  • cache-dit (DBCache)v1.5.6from November 20, 2026new release

    1.5–3× at nearly the same quality

    Reuses block outputs between diffusion steps. Supports most DiT models.

    How to enable: pip install cache-dit; one line cache_dit.enable_cache(pipe) for diffusers.

    Earlier generations (2)
    • MagCache — Skips steps by magnitude of change. More accurate than TeaCache, no calibration.
    • TeaCache — The first popular diffusion step cache. Works, newer methods are more accurate.
  • Nunchaku (SVDQuant 4-bit)v1.2.1from January 25, 2026

    ≈3× faster and 3.5× less memory

    4-bit versions of popular image models with almost no quality loss (FP4 on RTX 50).

    How to enable: Ready weights for supported models; in ComfyUI use ComfyUI-nunchaku.

  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. Krea-2-Rawfamily root
  2. Krea-2-Turbo
  3. 1125-Krea2Turbo-Asian-Utopian

New posts about versions of Krea-2-Raw (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face

About the model

hinablue/1125-Krea2Turbo-Asian-Utopian
Author
hinablue
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

1031

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments