[CVPR 2026] 🎨 PosterOmni:
Generalized Artistic Poster Creation via Task Distillation and Unified Reward Feedback

👥 Authors

Sixiang Chen^1,2*, Jianyu Lai^1,2*, Jialin Gao²*, Hengyu Shi²*, Zhongying Liu²*, Tian Ye¹, Junfeng Luo², Xiaoming Wei², Lei Zhu^1,3†

¹HKUST (GZ)   ²Meituan   ³HKUST

*Core Contribution, †Corresponding Author

---

💡 We also have other text-to-poster generation that may interest you ✨.

[ICLR 2026] PosterCraft: Rethinking High-Quality Aesthetic Poster Generation in a Unified Framework
Sixiang Chen*, Jianyu Lai*, Jialin Gao*, Tian Ye, Haoyu Chen, Hengyu Shi, Shitong Shao, Yunlong Lin, Song Fei, Zhaohu Xing, Yeying Jin, Junfeng Luo, Xiaoming Wei, Lei Zhu

---

🔥 News & Updates

• 🚀 [2026.02] Our gradio demo and inference code are now available!
• 📊 [2026.02] We have released PosterOmni weights on HuggingFace.
• 📄 [2026.02] Our paper is now available on arXiv.

🚀 Quick Start

🔧 Installation

git clone https://github.com/MeiGen-AI/PosterOmni.git
cd PosterOmni

conda create -n posteromni python=3.11 -y
conda activate posteromni

# Install latest diffusers from source (required)
pip install git+https://github.com/huggingface/diffusers

# Install other dependencies
pip install -r requirements.txt

---

🚀 Inference

PosterOmni supports high-quality poster generation and 6 editing/creation tasks:

• poster_rescaling
• poster_filling
• poster_extending
• poster_layout_driven_generation
• poster_style_driven_generation
• poster_id_driven_generation

Basic Usage

python posteromni_inference.py \
  --task poster_rescaling \
  --image_path asset/rescale_1.jpg \
  --prompt "Rescale image to 16:9." \
  --base_model_path "Qwen-Image-Edit" \
  --transformer_path "MeiGen-AI/PosterOmni_v1" \
  --output_path output_rescale.png \
  --steps 40 \
  --cfg 4.0 \
  --guidance 1.0 \
  --seed -1

---

⚙️ Arguments

Argument	Description	Required	Default
--task	Task to perform. Options: poster_rescaling, poster_filling, poster_extending, poster_layout_driven_generation, poster_style_driven_generation, poster_id_driven_generation	✅	-
--image_path	Path to the input image(s). Supports multiple paths (space-separated), especially for ID-driven generation.	✅	-
--prompt	Text prompt. For rescaling, must include target ratio like to 16:9	✅	-
--base_model_path	Base model path/name that provides the full pipeline (e.g., Qwen-Image-Edit).	✅	-
--transformer_path	PosterOmni transformer weights (HF repo or local path). If not set, uses base model transformer	❌	None
--lora_path	Optional LoRA weights (.safetensors)	❌	None
--output_path	Output image file path	❌	output.png
--steps	Number of inference steps	❌	40
--cfg	CFG scale (true_cfg_scale)	❌	4.0
--guidance	Guidance scale (guidance_scale)	❌	1.0
--seed	Random seed. Use -1 for random	❌	-1
--no_resize	Disable auto-resize to ~1024×1024 pixels	❌	False

Note (Rescaling prompt format): please include to W:H, e.g. Rescale image to 1:1 or Rescale image to 16:9.

---

🧪 Task Examples

1) Poster Rescaling

python posteromni_inference.py \
  --task poster_rescaling \
  --base_model_path "Qwen-Image-Edit" \
  --transformer_path "MeiGen-AI/PosterOmni_v1" \
  --image_path asset/rescale_1.jpg \
  --prompt "Rescale image to 16:9." \
  --output_path output_rescale.png \
  --steps 40 \
  --cfg 4.0 \
  --guidance 1.0 \
  --seed -1

---

2) Layout-driven Generation

python posteromni_inference.py \
  --task poster_layout_driven_generation \
  --base_model_path "Qwen-Image-Edit" \
  --transformer_path "MeiGen-AI/PosterOmni_v1" \
  --image_path asset/layout_1.jpg \
  --prompt "Refer to the layout of this poster and create a new poster featuring a large beige speaker cabinet filled with plush toys. Next to it, place a metal high stool and a paintbrush with a wooden handle. On the right side, include a cluster of blooming light purple 3D-printed models. Add the text \"Smart Space: Innovative Living\" at the top and \"Collaborative Exploration, Enhanced Experience\" at the bottom." \
  --output_path output_layout.png \
  --steps 40 \
  --cfg 4.0 \
  --guidance 1.0 \
  --seed -1

---

3) Style-driven Generation

python posteromni_inference.py \
  --task poster_style_driven_generation \
  --base_model_path "Qwen-Image-Edit" \
  --transformer_path "MeiGen-AI/PosterOmni_v1" \
  --image_path asset/style_1.png \
  --prompt "参考这张图的风格生成一张全新的海报，主体一只猫，标题\"smiley cat\"" \
  --output_path output_style.png \
  --steps 40 \
  --cfg 4.0 \
  --guidance 1.0 \
  --seed -1

---

4) ID-driven Generation (Multi-image)

Tip: You can pass one or more reference images after --image_path (space-separated).

python posteromni_inference.py \
  --task poster_id_driven_generation \
  --base_model_path "Qwen-Image-Edit" \
  --transformer_path "MeiGen-AI/PosterOmni_v1" \
  --image_path asset/id_1.jpg \
  --prompt "一张海报上头上顶西瓜皮的黄色卡通玩偶站在广袤的草地上，看着树上的苹果。标题\"水豚噜噜的凝视\"" \
  --output_path output_id.png \
  --steps 40 \
  --cfg 4.0 \
  --guidance 1.0 \
  --seed -1

---

(Optional) Run with GPU

CUDA_VISIBLE_DEVICES=2 python posteromni_inference.py \
  --task poster_rescaling \
  --image_path asset/rescale_1.jpg \
  --prompt "Rescale image to 16:9." \
  --base_model_path "Qwen-Image-Edit" \
  --transformer_path "MeiGen-AI/PosterOmni_v1" \
  --output_path output_rescale.png

---

🧪 Demo

python demo_gradio.py

---

🌟 What is PosterOmni?

PosterOmni is a unified image-to-poster framework that covers both:

• Poster Local Editing: Rescaling, Filling, Extending, Identity-driven
• Poster Global Creation: Layout-driven, Style-driven
• Unified Framework: one interface to handle multi-task poster generation & editing.

---

📚 PosterOmni Dataset

We introduce a unified data suite with PosterOmni-200K (training) and PosterOmni-Bench (evaluation) for image-to-poster generation. PosterOmni-200K contains 200K+ paired samples covering six tasks—local editing (Rescaling, Filling, Extending, Identity-driven) and global creation (Layout-driven, Style-driven)—and spans six poster themes: Products, Food, Events/Travel, Nature, Education, Entertainment.
PosterOmni-Bench provides 540 Chinese and 480 English prompts, evenly distributed across the same six themes for consistent evaluation across tasks.

🏗️ Model Architecture

PosterOmni is trained with a four-stage workflow (see figure below) that progressively unifies local editing and global creation capabilities.

🧩 Stage 1: Task-specific SFT

Trains edit experts (Rescale/Fill/Extend/Identity) and creation experts (Layout/Style) with task-specific supervision, establishing strong specialized abilities for each poster task.

🔄 Stage 2: Task Distillation

Distills knowledge from task experts into a single multi-task model, enabling consistent behavior across both local and global tasks.

⭐ Stage 3: PosterOmni Reward Training

Learns a unified reward signal to evaluate poster results across tasks, emphasizing text fidelity, visual consistency, and overall poster quality.

🚀 Stage 4: Omni-Edit Reinforcement Learning

Further aligns the unified model with the reward signal through reinforcement learning, improving robustness and quality across diverse poster editing and creation scenarios.

---

📊 Performance Benchmarks

---

💾 Model Zoo

Model	Type	Download
PosterOmni-v1.0	unified	🤗 https://huggingface.co/MeiGen-AI/PosterOmni_v1

---

🙏 Acknowledgments

• 🏛️ Thanks to our affiliated institutions for their support.
• 🤝 Special thanks to the open-source community for inspiration.

---

📬 Contact

For any questions or inquiries, please reach out to us:

• Sixiang Chen: schen691@connect.hkust-gz.edu.cn
• Jianyu Lai: jlai218@connect.hkust-gz.edu.cn
• Jialin Gao: gaojialin04@meituan.com
• Hengyu Shi: qq1842084@gmail.com
• Zhongying Liu: liuzhongying@meituan.com

---

📝 Citation

If you find PosterOmni useful for your research, please cite our paper:

@article{chen2026posteromni,
  title={PosterOmni: Generalized Artistic Poster Creation via Task Distillation and Unified Reward Feedback},
  author={Chen, Sixiang and Lai, Jianyu and Gao, Jialin and Shi, Hengyu and Liu, Zhongying and Ye, Tian and Luo, Junfeng and Wei, Xiaoming and Zhu, Lei},
  journal={arXiv preprint arXiv:2602.12127},
  year={2026}
}