README.md

metadata

dataset_info:
  features:
    - name: prompt
      dtype: string
    - name: image
      dtype: image
    - name: tag
      dtype: string
    - name: claude-3-haiku-expanded
      dtype: string
    - name: clip-description
      dtype: string
    - name: florence-2-caption
      dtype: string
    - name: gemini-1.5-flash-description
      dtype: string
    - name: gemma3-4b-description
      dtype: string
    - name: gpt-4o-mini-expanded
      dtype: string
    - name: kosmos2-caption
      dtype: string
    - name: llama-3.1-405b-expanded
      dtype: string
    - name: llava-1.5-7b-hf-description
      dtype: string
    - name: llava-phi3-description
      dtype: string
    - name: mistral-small-3.2-24b-instruct-2506-description
      dtype: string
    - name: moondream2-long-description
      dtype: string
    - name: qwen2-72b-expanded
      dtype: string
  splits:
    - name: train
      num_bytes: 6924257181.452
      num_examples: 8551
  download_size: 6080196160
  dataset_size: 6924257181.452
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
task_categories:
  - text-to-image
  - image-to-text
language:
  - en
tags:
  - preference
  - caption
  - midjourney
  - image
size_categories:
  - 1K<n<10K

MJ Showcase 2024

This is a dataset of the top voted creations which were manually collected daily between May and August 2024.

150 most frequent words used in the prompts

〽️ Stats and Data Composition

1. Described: Both Image and Prompt as input
2. Expanded: Only Prompt as input
3. Captioned: Only Image as input

Model-Technique	Rows	Incomplete
llava-1.5-7b-hf-described	8551	0
clip-described	8551	0
llama-3.1-405b-expanded	8551	0
claude-3-haiku-expanded	8551	0
gemma3-4b-described	8551	0
llava-phi3-described	8551	0
gemini-1.5-flash-described	8551	38
moondream2-long-described	8551	0
qwen2-72b-expanded	8551	0
kosmos2-captioned	8551	0
gpt-4o-mini-expanded	8551	0
florence-2-captioned	8551	0
mistral-small-3.2-24b-instruct-2506-described	8551	0

Top Tags

1. --ar

• --16:9 : 1,339 times (8.67%)
• --9:16 : 1,229 times (7.95%)
• --2:3 : 911 times (5.90%)

2. --stylize

• --250 : 688 times (4.45%)
• --750 : 626 times (4.05%)
• --50 : 273 times (1.77%)

3. --v

• --6.1 : 1,551 times (10.04%)
• --6 : 211 times (1.37%)
• --5.2 : 62 times (0.40%)

4. --style

• --raw : 1,298 times (8.40%)

5. --chaos

• --10 : 155 times (1.00%)
• --20 : 93 times (0.60%)
• --30 : 89 times (0.58%)

The --personalize tag was filtered out.

🎯 Uses

For learning prompting styles and techniques, preference training of text to image models and captioning.

Please refer to the usage policies of the respective models and adhere to their policies.

⚠️ Limitations

You may find prompts which dont have any meaningful words.

The missing 38 descriptions from gemini are refusals.

This dataset is biased towards the preferences of those who actively vote on the images.

⚖️ License

This dataset is licensed under the Creative Commons Attribution 4.0 International License.

📝 Citation

If you utilize this dataset, please cite it using the BibTeX provided.

@misc {sb_2025,
    author       = { SB },
    title        = { mj-showcase-8K },
    year         = 2025,
    url          = { https://huggingface.co/datasets/shb777/mj-showcase-8K },
    publisher    = { Hugging Face }
}