PromptMoG: Prompt Embedding MoG Sampling

A plug-and-play framework for enhancing diversity in long-prompt image generation

Bo-Kai Ruan, Teng-Fang Hsiao, Ling Lo, Yi-Lun Wu, Hong-Han Shuai

CogView4
+PromptMoG (Ours)
+CADS
+DiverseFlow

Qwen-Image
+PromptMoG (Ours)
+CADS
+DiverseFlow

Note

We provide the python package for LPD-Bench evaluation. Please check the lpd-eval branch for more details.

📦 Installation

# Clone the repository
git clone https://github.com/basiclab/PromptMog.git --depth 1 && cd PromptMoG

# Install the dependencies
uv sync

# load environment
source .venv/bin/activate

# If users want to compute the dataset statistics, download the spacy model
uv run spacy download en_core_web_sm

🗃️ Data Preparation

We have provided the filtered dataset in data/lpd_bench. Users can also follow the following steps to generate the dataset themselves.

Creating LPD-Bench

To generate the long prompts similar to LPD-Bench, run the following command:

# Step 1: Generate a pool of long prompts
python misc/dataset_gen/generate_long_prompt.py --num-prompts-for-topic 60

# Step 2: Filter the prompts
python misc/dataset_gen/post_process_data.py --data-root data/lpd_bench --num_prompts_per_topic 60 --num_remain_per_topic 40

The outputs will be saved to data/lpd_bench/filtered.

Users can also run the statistics of the dataset by running the following command:

python misc/dataset_gen/data_statistics.py --data-root-dir data --plot 

The results will be saved to assets/dataset_statistics.pdf.

Creating LPD-Bench-Rewritten

To rewrite the long prompts, run the following command:

python misc/rewrite_long_prompt.py \
    --data_folder data/lpd_bench/filtered \
    --output_folder data/lpd_bench/rewritten \
    --num_variants 10 \
    --model gpt-4o \
    --workers 8

The outputs will be saved to data/lpd_bench/rewritten.

🚀 Usage

Diversity Test

# Generate images
./scripts/gen_image.sh \
    --dataset_type long \
    --prompt_root_dir data/lpd_bench \
    --output_root_dir outputs/long_prompt
./scripts/gen_image.sh \
    --dataset_type short \
    --prompt_root_dir data/lpd_bench \
    --output_root_dir outputs/short_prompt_1 \
    --first_top 1
./scripts/gen_image.sh \
    --dataset_type short \
    --prompt_root_dir data/lpd_bench \
    --output_root_dir outputs/short_prompt_3 \
    --first_top 3

# Score the diversity
./scripts/scoring_diversity.sh --output_root_dir outputs/long_prompt
./scripts/scoring_diversity.sh --output_root_dir outputs/short_prompt_1
./scripts/scoring_diversity.sh --output_root_dir outputs/short_prompt_3

Prompt-MoG

./scripts/gen_image.sh \
    --dataset_type long \
    --model_type pmog \
    --prompt_root_dir data/lpd_bench \
    --output_root_dir outputs/pmog

Chunking

./scripts/gen_image.sh \
    --dataset_type long \
    --model_type chunk \
    --prompt_root_dir data/lpd_bench \
    --output_root_dir outputs/chunk_prompt

Prompt Rewriting

./scripts/gen_image.sh \
    --dataset_type rewritten \
    --prompt_root_dir data/lpd_bench \
    --output_root_dir outputs/rewritten_prompt \
    --model_type short

./scripts/scoring_diversity.sh --output_root_dir outputs/rewritten_prompt
./scripts/scoring_lpd.sh --output_root_dir outputs/rewritten_prompt

GenEval

./scripts/gen_image.sh \
    --dataset_type gen_eval \
    --prompt_root_dir data/geneval \
    --output_root_dir outputs/gen_eval_prompt

Ablation Study

Model	Default Gamma	Default Num Mode	Default Sigma
Flux	0.6	50	0.25
Qwen	0.85	50	0.25

Exploring the gamma

for gamma in 1.1 0.85 0.6 0.35 0.1; do
  ./scripts/ablation/gen_and_scoring_ablation.sh \
      --prompt_root_dir data/lpd_bench \
      --output_root_dir outputs/ablation_gamma_flux_${gamma}_qwen_${gamma} \
      --dataset_type long \
      --model_type pmog \
      --partial_num 8 \
      --flux_gamma ${gamma} \
      --qwen_gamma ${gamma}
done

Exploring the sigma

for sigma in 1.0 0.75 0.5 0.25 0.0; do
  ./scripts/ablation/gen_and_scoring_ablation.sh \
      --prompt_root_dir data/lpd_bench \
      --output_root_dir outputs/ablation_sigma_flux_${sigma}_qwen_${sigma} \
      --dataset_type long \
      --model_type pmog \
      --partial_num 8 \
      --flux_sigma ${sigma} \
      --qwen_sigma ${sigma}
done

Exploring the mode

for mode in 100 75 50 25 1; do
  ./scripts/ablation/gen_and_scoring_ablation.sh \
      --prompt_root_dir data/lpd_bench \
      --output_root_dir outputs/ablation_mode_flux_${mode}_qwen_${mode} \
      --dataset_type long \
      --model_type pmog \
      --partial_num 8 \
      --flux_num_mode ${mode} \
      --qwen_num_mode ${mode}
done

📝 Citation

@article{ruan2025promptmog,
  title={{PromptMoG}: Enhancing Diversity in Long-Prompt Image Generation via
Prompt Embedding Mixture-of-Gaussian Sampling},
  author={Ruan, Bo-Kai and Hsiao, Teng-Fang and Lo, Ling and Wu, Yi-Lun and Shuai, Hong-Han},
  journal={arXiv preprint arXiv:2511.20251},
  year={2025}
}