Music Cover CLIP Embeddings for ~3.5 Million Album Covers

CLIP ViT-L/14 embeddings for approximately 3.5 million music album covers combine image and text features. Each row contains a 1536-dimensional, L2-normalized vector concatenating the cover image embedding and a text embedding of the format '{title} by: {artist}'. The dataset was created by the author 'dyslexi' and was last updated on 2026-04-30.

Use Cases

• Music retrieval based on album cover visual similarity using the image embeddings.
• Artist or title search based on the text embeddings derived from '{title} by: {artist}'.
• Training or evaluating multimodal models for music-related tasks using the combined 1536-dimensional embeddings.
• Analyzing the relationship between album art and musical metadata via the fused image-text representation.

Strengths

• Contains embeddings for approximately 3.5 million music album covers.
• Provides a unified 1536-dimensional vector for each cover, combining image and text information from a known model (CLIP ViT-L/14).
• All vectors are L2-normalized, facilitating similarity search via cosine distance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Collection Method

Embeddings generated using the openai/clip-vit-large-patch14 model.

Freshness

Last updated 2026-04-30 11:40:11; freshness should be verified.