Movie Posters 100K ControlNet
A comprehensive visual dataset of annotated movie posters for training layout-guided generation models like ControlNet.
10,000 high-resolution images + OCR annotations, approximately 42.5 GB, Parquet format
Apache 2.0
Description
Movie Posters 100K ControlNet is a corpus of 10,000 high-resolution movie posters, enriched with layout annotations and captions. Each image is associated with a title and one or more genres, forming a useful caption for content generation. The visual structures were extracted via PaddleOCR and can be used to train conditional models like ControlNet.
What is this dataset for?
- Train image generation models from layouts (layout2image)
- Use ControlNet to generate custom posters from a canvas
- Analyze the graphic structure of posters for classification or automated design projects
Can it be enriched or improved?
Yes, this dataset can be enriched with additional metadata (movie summaries, dates, directors), standardization of OCR annotations, or by adding cultural or linguistic variations. It is also possible to cross it with other movie databases to extend its use.
🔎 In summary
🧠 Recommended for
- Visual AI researchers
- Generative artists
- layout2image projects
🔧 Compatible tools
- ControlNet
- Stable Diffusion
- PaddleOCR
- Diffusers
💡 Tip
For precise control, use OCR layouts as an input mask in your build prompts.
Frequently Asked Questions
Does this dataset include the full texts of the movies?
No, it only includes poster images, genres and titles as captions, and their extracted layout.
Is it suitable for training a ControlNet model?
Yes, it is one of its main uses: to provide a visual base with conditioned layout to generate new posters.
Can we use this dataset to classify movie genres?
Yes, captions include genres, which makes it possible to train a visual or multimodal classifier according to graphic appearance.
