MegaStyle: Constructing Diverse and Scalable Style Dataset via Consistent Text-to-Image Style Mapping

MegaStyle is a novel and scalable data curation pipeline that first explores consistent T2I style mapping ability from current large generative models to construct intra-style consistent, inter-style diverse and high-quality style dataset.

Your star is our fuel! We're revving up the engines with it! Check out our project page for more visual results!

News

[2026/4/23] 🔥 We release a Gradio demo and ComfyUI custom nodes (with a ready-to-use workflow) for style transfer using MegaStyle-FLUX.
[2026/4/22] 🔥 Thanks to @olfronar's contribution! The style score computation using MegaStyle-Encoder is now available on HF space.
[2026/4/21] 🔥 We release the training/inference codes, models and dataset of MegaStyle!!!

TODO List

A more diverse and larger-scale style dataset.

MegaStyle-1.4M

MegaStyle-1.4M is a large-scale style dataset built through a scalable pipeline that leverages consistent text-to-image style mapping of Qwen-Image. It combines 170K curated style prompts with 400K content prompts to generate 1.4M high-quality images that share strong intra-style consistency while covering diverse fine-grained styles.

Get Started

Trained on MegaStyle1.4M, we introduce MegaStyle-FLUX and MegaStyle-Encoder for generalizable style transfer and reliable style similarity measurement.

Clone the Repository

git clone git@github.com:Tencent/MegaStyle.git
cd ./MegaStyle

Environment Setup

conda create -n megastyle python==3.10
conda activate megastyle
pip install diffsynth==1.1.8

Downloading Checkpoints

Download the pretrained models of SigLIP and FLUX.1-dev.
Download the models into ./models/.

Running Inference

For image style transfer, we provide 50 reference style images from StyleBench in ./ref_styles:

python inference.py --ckpt_path models/megastyle_flux.safetensors --ref_path ./ref_styles

For computing style score:

python style_score.py --ckpt_path models/megastyle_encoder.pth --real_image_path <path/to/image.png> --fake_image_path <path/to/image.png>

Gradio Demo

An interactive web UI is provided via gradio_demo.py. Install Gradio and launch:

pip install gradio
python gradio_demo.py --ckpt_path models/megastyle_flux.safetensors --ref_path ./ref_styles

Then open http://localhost:8080 in your browser. Upload a reference style image, type a content prompt, and click Generate. Common options:

python gradio_demo.py \
    --ckpt_path models/megastyle_flux.safetensors \
    --ref_path ./ref_styles \
    --server_name 0.0.0.0 --server_port 8080 [--share]

ComfyUI Custom Nodes

Custom nodes live in ./comfyui/ and, together with the shipped workflow_megastyle.json, make MegaStyle available as a drop-in graph inside ComfyUI. The exposed nodes mirror a standard Flux workflow:

Models Loader — loads FLUX.1-dev into a FluxImagePipeline.
MegaStyle LoRA Loader — patches the MegaStyle-FLUX LoRA onto the DiT.
Reference Style — LoadImage input for the style reference.
Text Encode — CLIP + T5 prompt encoding.
VAE Encode — encodes the reference style image into latents.
Flow Matching Scheduler — denoise loop with enable_shift_rope=True.
VAE Decode — decodes latents back to an image.
Save Image — writes results to output/MegaStyle/.

1. Clone & install ComfyUI (skip if you already have one)

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
conda activate megastyle            # reuse the MegaStyle env (needs diffsynth==1.1.8)
pip install -r requirements.txt
cd ..

2. Register the MegaStyle node package

From the MegaStyle repo root (so that flux_image_mega.py stays importable):

# Option A (recommended): symlink the comfyui package directly.
ln -s "$(pwd)/comfyui" /path/to/ComfyUI/custom_nodes/MegaStyle

# Option B: symlink the whole repo, then drop a one-line shim.
ln -s "$(pwd)" /path/to/ComfyUI/custom_nodes/MegaStyle
echo 'from .comfyui import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS' \
    > /path/to/ComfyUI/custom_nodes/MegaStyle/__init__.py

On first launch the package will also:

copy comfyui/workflow_megastyle.json to ComfyUI/user/default/workflows/MegaStyle.json so it shows up in the Workflows side panel;
symlink ref_styles/*.jpg into ComfyUI/input/ so the default LoadImage node resolves 00.jpg out of the box.

Disable with MEGASTYLE_AUTO_INSTALL_WORKFLOW=0 / MEGASTYLE_AUTO_INSTALL_REFS=0. If auto-discovery of the ComfyUI root fails, set MEGASTYLE_COMFY_ROOT=/path/to/ComfyUI.

3. Launch & run

cd /path/to/ComfyUI
python main.py --listen 0.0.0.0 --port 8080

Open http://localhost:8080, pick the MegaStyle workflow from the Workflows panel, then click Queue Prompt. The default lora_path is models/megastyle_flux.safetensors (resolved relative to the MegaStyle repo root); set it to an absolute path if you keep the checkpoint elsewhere. See ./comfyui/README.md for the wiring diagram and advanced options (CFG, custom negative prompts, etc.).

Training

To train a style transfer model with paired supervision, please download our style dataset, MegaStyle1.4M, and start training with:

bash FLUX.1-dev.sh # FLUX.1-dev-npu.sh for npu

License and Citation

All assets and code are under the license unless specified otherwise.

If this work is helpful for your research, please consider citing the following BibTeX entry.

@article{gao2026megastyle,
  title={MegaStyle: Constructing Diverse and Scalable Style Dataset via Consistent Text-to-Image Style Mapping},
  author={Gao, Junyao and Liu, Sibo and Li, Jiaxing and Sun, Yanan and Tu, Yuanpeng and Shen, Fei and Zhang, Weidong and Zhao, Cairong and Zhang, Jun},
  journal={arXiv preprint arXiv:2604.08364},
  year={2026}
}

Acknowledgements

The code is built upon DiffSynth-Studio.